About Image Processing Technology
Image recognition technology refers to the technology in which a computer analyzes a digital image and identifies objects, people, scenery, etc. in the image. The algorithms used in these technologies can be broadly classified into the following categories.
- Feature Extraction Algorithms: Algorithms that extract characteristic parts from an image. For example, edge detection, color information, shape information, etc. are extracted.
- Classification algorithms: These algorithms classify objects, people, landscapes, etc. using image features. Typical algorithms include support vector machines (SVM), decision trees, random forests, and neural networks.
- Deep learning (DNN) algorithms: These algorithms use multi-layer neural networks and are capable of advanced image recognition. Typical examples include convolutional neural networks (CNN), recurrent neural networks (RNN), and deep learning.
Among these, DNN algorithms such as CNNs are commonly used because feature extraction and classification can be achieved simultaneously, and high accuracy can be obtained. However, since DNN requires a large amount of training data, approaches combining other algorithms are also being considered when only a small amount of data is available.
This image recognition technology is used in a wide variety of fields, including security surveillance, medical imaging, automated driving technology, robotics, and image retrieval. The following are typical examples of their applications.
- Security Surveillance: This technology is used in systems that analyze video from surveillance cameras to detect suspicious activity or anomalies. This could be, for example, facial recognition of people on surveillance cameras or technology to identify specific objects.
- Medical Imaging: Medical images are analyzed to detect diseases and abnormalities. For example, this could be used to diagnose lung cancer or stroke from X-rays or CT images.
- Automated driving technology: This technology is used to detect roads, obstacles, pedestrians, etc. by analyzing information from cameras and sensors installed in automobiles to realize automated driving.
- Robotics: Robots are equipped with cameras and sensors to understand their surroundings and automate tasks. This could be, for example, recognizing and sorting parts in a factory or guiding logistics robots.
- Image retrieval: This is used to analyze images on the Internet and retrieve images that match keywords. This can be used, for example, to analyze product images and search for products on online shopping sites.
In this section, we discuss the theory and various practical applications of these image recognition techniques, including approaches other than deep learning techniques.
Implementation
An image recognition system will be a technology in which a computer analyzes images and automatically identifies objects and features contained in them. This system is implemented by combining various artificial intelligence algorithms and methods, such as image processing, pattern recognition, machine learning, and deep learning. This section describes the steps for building this image recognition system and their specific implementation.
In image information processing, preprocessing has a significant impact on model performance and convergence speed, and is an important step in converting image data into a form suitable for the model. The following describes preprocessing methods for image information processing.
The main approaches to using artificial intelligence techniques to extract emotions include (1) natural language processing, (2) speech recognition, (3) image recognition, and (4) biometric analysis. These methods are combined with algorithms such as machine learning and deep learning, and are basically detected using large amounts of training data. Approaches that combine different modalities (text, voice, images, biometric information, etc.) to comprehensively understand emotions are also more accurate methods.
Various models for emotion recognition have been proposed, as described in “Emotion recognition, Buddhist philosophy and AI”. In addition, a number of AI technologies such as speech recognition, image recognition, natural language processing and bioinformation analysis have been used to extract emotions. This section describes the details of these technologies.
The Frank-Wolfe method is a numerical algorithm for solving non-linear optimisation problems, proposed by Marguerite Frank and Philippe Wolfe in 1956. The Frank-Wolfe method is also related to linear programming problems and can be applied to continuous optimisation problems. However, its convergence speed may be slower than that of general optimisation algorithms, and therefore other efficient algorithms may be preferred for high-dimensional problems. The Frank-Wolff method is useful in large-scale and constrained optimisation problems and is widely used in machine learning, signal processing and image processing. The Frank-Wolff method is also often used in combination with other optimisation methods.
CNN (Convolutional Neural Network) is a deep learning model mainly used for computer vision tasks such as image recognition, pattern recognition, and image generation. This section provides an overview of CNNs and implementation examples.
Contrastive Predictive Coding (CPC) is a representation learning technique used to learn semantically important representations from audio and image data. This method is a form of unsupervised learning, in which representations are learned by contrasting different observations in the training data.
DenseNet (Densely Connected Convolutional Network) was proposed in 2017 by Gao Huang, Zhuang Liu, Kilian Q. Weinberger, and Laurens van der Maaten in “Overview of CNN DenseNet improves the efficiency of deep network training by introducing “dense” connections during convolutional neural network training, and reduces the gradient loss problem. and reducing the gradient loss problem.
ResNet is a deep convolutional neural network (CNN) architecture proposed by Kaiming He et al. in 2015, as described in “CNN Overview, Algorithms and Implementation Examples”. ResNet introduces innovative ideas and approaches that have achieved phenomenal performance in computer vision tasks.
GoogLeNet is a convolutional neural network (CNN) architecture described in Google’s 2014 “CNN Overview and Algorithms and Examples of Implementations”. This model achieved state-of-the-art performance in computer vision tasks such as the ImageNet Large Scale Visual Recognition Challenge (ILSVRC), and GoogLeNet is known for its unique architecture and modular structure. GoogLeNet is known for its unique architecture and modular structure.
VGGNet (Visual Geometry Group Network) is a convolutional neural network (CNN) model developed in 2014 and described in “CNN Overview, Algorithms, and Examples of Implementations” that has achieved high performance in computer vision tasks. VGGNet was proposed by researchers in the Visual Geometry Group at the University of Oxford.
AlexNet is a deep learning model proposed in 2012 that represents a breakthrough in computer vision tasks. Convolutional Neural Networks (CNNs), which are primarily used for image recognition tasks.
The multi-class object detection model will be a machine learning model for performing the task of simultaneously detecting several objects of different classes (categories) in an image or video frame and enclosing the locations of these objects with bounding boxes. Multiclass object detection is used in important applications in computer vision and object recognition, and has been applied in various fields such as automated driving, surveillance, robotics, and medical image analysis.
Adding a head for refining position information (e.g., regression head) to the object detection model is a very important approach to improve the performance of object detection. This head helps to adjust the coordinates and size of the object bounding box to more accurately position the detected object.
Detecting small objects in image detection is generally a difficult task. Because small objects have few pixels, their features may be obscured and difficult to capture with normal resolution feature maps, making the use of image pyramids and high-resolution feature maps an effective approach in such cases.
Object detection technology involves the automatic detection of specific objects or objects in an image or video and their location. Object detection is an important application of computer vision and image processing and is applied to many real-world problems. This section describes various algorithms and implementation examples for this object detection technique.
Haar Cascades is a feature-based algorithm for object detection, and Haar Cascades is widely used for computer vision tasks, especially face detection. This section provides an overview of this Haar Cascades and its algorithm and implementation.
Intersection over Union (IoU) is one of the evaluation metrics used in computer vision tasks such as object detection and region suggestion, and is an indicator of the overlap between the predicted bounding box and the true bounding box.
Anchor boxes in object detection is a concept widely used in convolutional neural network (CNN)-based object detection algorithms, where anchor boxes are used to represent candidate object regions at multiple locations and scales in an image.
Selective Search is one of the candidate region suggestion methods for object detection used in the field of computer vision and object detection, where object detection is the task of locating objects in an image, which is one of the key applications of computer vision. Selective Search helps object detection models to suggest regions where objects are likely to be present.
The EdgeBoxes algorithm is one of the candidate region suggestion methods for object detection. This method is used to locate potential objects in an image and efficiently and quickly suggests regions where objects are likely to be present.
Proposal networks are a type of neural network used mainly in the fields of computer vision and image processing, especially for object detection and region proposal (object proposal) tasks. A proposal network is a model for proposing a region of interest (an object or an area in which an object is present) from an input image.
Histogram of Oriented Gradients (HOG) is a feature extraction method used for object detection and recognition in the fields of computer vision and image processing. The principle of HOG is to capture information on edges and gradient directions in an image and represent object features based on this information. This section provides an overview of HOG, its challenges, various algorithms, and implementation examples.
Cascade Classifier is one of the pattern recognition algorithms used in object detection tasks. Cascade classifiers have been developed to achieve fast object detection, and in particular, the Haar Cascades form is widely known and used mainly for tasks such as face detection. This section provides an overview of this cascade classifier, its algorithms, and examples of implementations.
R-CNN (Region-based Convolutional Neural Networks) is an approach to utilize deep learning in object detection tasks. neural networks (CNNs) to predict object classes and bounding boxes, and R-CNNs have shown very good performance in object detection tasks. This paper describes an overview of this R-CNN, its algorithm and implementation examples.
Faster Region-based Convolutional Neural Networks (Faster R-CNN) is one of a series of deep learning models that provide fast and accurate results in object detection tasks. Convolutional Neural Networks (R-CNNs)), and represents a major advance in the field of object detection, solving the problems of previous architectures called R-CNNs. This section provides an overview of this Faster R-CNN, its algorithms, and examples of implementations.
YOLO (You Only Look Once) is a deep learning-based algorithm for real-time object detection tasks. YOLO will be one of the most popular models in the fields of computer vision and artificial intelligence.
SSD (Single Shot MultiBox Detector) is one of the deep learning based algorithms for object detection tasks.
Mask R-CNN (Mask Region-based Convolutional Neural Network) is a deep learning-based architecture for object detection and object segmentation (instance segmentation), in which the location of each object is not only enclosed in a bounding box It has the ability to segment objects at the pixel level within an object as well as surround it, making it a powerful model for combining object detection and segmentation.
EfficientDet will be one of the computer vision models with high performance in the object detection task; EfficientDet is designed to balance the efficiency and accuracy of the model, and will provide superior performance with less computational resources.
RetinaNet is a deep learning-based architecture that performs well in object detection tasks by predicting the location of object bounding boxes and simultaneously estimating the probability of belonging to each object class. This architecture is based on an approach known as Single Shot Detector (SSD), which is also described in “Overview of SSD (Single Shot MultiBox Detector), Algorithms, and Examples of Implementations,” but it is more suitable for finding smaller or more difficult objects than a typical SSD. However, it performs better than the general SSD in detecting small or difficult-to-find objects.
Anchor Boxes and high Intersection over Union (IoU) thresholds play an important role in the object detection task of image recognition. The following sections discuss adjustments related to these elements and the detection of dense objects.
Diffusion Models are a class of generative models that perform well in tasks such as image generation and data repair. These models are generated by “diffusing” the original data in a series of steps.
DDIM (Diffusion Denoising Score Matching) is a method for removing noise from images. This approach uses a diffusion process to remove noise, combined with a statistical method called score matching. In this method, a noise image is first generated by adding random noise to the input image, and then the diffusion process is applied to these noise images as input to remove the noise by smoothing the image structure. Score matching is then used to learn the probability density function (PDF) of the noise-removed images. Score matching estimates the true data distribution by minimizing the difference between the gradient (score) of the denoised image and the gradient of the true data distribution, thereby more accurately recovering the true structure of the input image.
Denoising Diffusion Probabilistic Models (DDPMs) are probabilistic models used for tasks such as image generation and data completion, which model the distribution of images and data using a stochastic generative process.
Non-Maximum Suppression (NMS) is an algorithm used in computer vision tasks such as object detection, mainly for selecting the most reliable one from multiple overlapping bounding boxes or detection windows. It will be.
Stable Diffusion is a method used in the field of machine learning and generative modeling, and is an extension of the Diffusion Models described in “Overview, Algorithms, and Examples of Implementations of Diffusion Models,” which are known generative models for images and audio. Diffusion Models are known for their high performance in image generation and restoration, and Stable Diffusion expands on this to enable higher quality and more stable generation.
EfficientNet is one of the lightweight and efficient deep learning models and convolutional neural network (CNN) architectures.EfficientNet was proposed by Tan and Le in 2019 and was designed to optimize model size and It will be designed to achieve high accuracy while optimizing computational resources.
LeNet-5 (LeNet-5) is one of the most important historical neural network models in the field of deep learning and was proposed in 1998 by Yann Lecun, a pioneer in convolutional neural networks (CNN), as described in “CNN Overview and Algorithm and Implementation Examples. LeNet-5 was very successful in the handwritten digit recognition task and has contributed to the subsequent development of CNNs.
MobileNet is one of the most widely used deep learning models in the field of computer vision, and is a lightweight and efficient convolutional neural network (CNN) optimized for mobile devices developed by Google, as described in “CNN Overview, Algorithms and Implementation Examples”. MobileNet can be used for tasks such as image classification, object detection, and semantic segmentation, and offers superior performance, especially on resource-constrained devices and applications. It offers superior performance.
SqueezeNet is a lightweight, compact deep learning model and architecture for convolutional neural networks (CNNs), as described in “CNN Overview, Algorithms, and Implementation Examples. neural networks with small file sizes and low computational complexity, and is primarily suited for resource-constrained environments and devices.
U-Net is one of the deep learning architectures in image segmentation (the task of assigning each pixel of an image to a corresponding class), proposed in 2015, this network is particularly useful in the field of medical image processing and semantic segmentation.
Automatic machine learning (AutoML) refers to methods and tools for automating the process of designing, training, and optimizing machine learning models.AutoML is particularly useful for users with limited machine learning expertise or those seeking to develop efficient models, with the following main goals. This section provides an overview of this AutoML and examples of various implementations.
Similarity is a concept that describes the degree to which two or more objects or things have common features or properties and are considered similar to each other, and plays an important role in evaluating, classifying, and grouping objects in terms of comparison and relatedness. This section describes the concept of similarity and general calculation methods for various cases.
A segmentation network is a type of neural network that can be used to identify different objects or regions in an image on a pixel-by-pixel basis and divide them into segments (regions). It is mainly used in computer vision tasks and plays an important role in many applications because it can associate each pixel in an image to a different class or category. This section provides an overview of this segmentation network and its implementation in various algorithms.
The issue of small amount of data to be trained (small data) is a problem that appears in various tasks as a factor that reduces the accuracy of machine learning. Machine learning with small data can be approached in various ways, taking into account data limitations and the risk of overlearning. This section discusses the details of each approach and implementation examples.
Transfer learning, a type of machine learning, is a technique for applying a model or knowledge learned in one task to a different task. Transfer learning is usually useful when a new task requires little data or high performance. This section provides an overview of transfer learning and various algorithms and implementation examples.
Self-Supervised Learning is a type of machine learning and can be considered as a type of supervised learning. While supervised learning uses labeled data to train models, self-supervised learning uses the data itself instead of labels to train models. This section describes various algorithms, applications, and implementations of self-supervised learning.
There are open source tools such as text-generation-webui and AUTOMATIC1111 that allow codeless use of generation modules such as ChatGPT and Stable Diffusion. In this article, we describe how to use these modules for text generation and image generation.
Support Vector Machine (SVM) is a supervised learning algorithm widely used in pattern recognition and machine learning. is to find the best separating hyperplane between the classes in the feature vector space, which is determined to have the maximum margin with the data points in the feature space. The margin is defined as the distance between the separation hyperplane and the nearest data point (support vector), and in SVM, the optimal separation hyperplane can be found by solving the margin maximization problem.
This section describes various practical examples of this support vector machine and their implementation in python.
Robust Principal Component Analysis (RPCA) is a method for finding a basis in data, and is characterized by its robustness to data containing outliers and noise. This paper describes various applications of RPCA and its concrete implementation using pyhton.
LightGBM is a Gradient Boosting Machine (GBM) framework developed by Microsoft, which is a machine learning tool designed to build fast and accurate models for large data sets. Here we describe its implementation in pyhton, R, and Clojure.
This section provides an overview of python Keras and examples of its application to basic deep learning tasks (handwriting recognition using MINIST, Autoencoder, CNN, RNN, LSTM).
Sparse modeling is a technique that takes advantage of sparsity in the representation of signals and data. Sparsity refers to the property that non-zero elements in data or signals are limited to a very small portion. The purpose of sparse modeling is to efficiently represent data by utilizing sparsity, and to perform tasks such as noise removal, feature selection, and compression.
This section provides an overview of sparse modeling algorithms such as Lasso, compression estimation, Ridge regularization, elastic nets, Fused Lasso, group regularization, message propagation algorithms, dictionary learning, etc., as well as a description of the various algorithms used in image processing, natural language processing, recommendation, signal processing The paper describes the implementation of the algorithms in various applications such as image processing, natural language processing, recommendation, machine learning, signal processing, brain science, and so on.
The trace norm (or nuclear norm) is a type of matrix norm, which can be defined as the sum of the singular values of a matrix. It plays a particularly important role in matrix low-rank approximation and matrix minimisation problems.
The Frobenius norm is a type of matrix norm, defined as the square root of the sum of squares of the elements of a matrix. This means that the Frobenius norm of the matrix \( A \), \( ||A||_F \), is given by the following equation.
\[ ||A||_F = \sqrt{\sum_{i=1}^m \sum_{j=1}^n |a_{ij}|^2} \]
Where ὅ( A = [a_{ij}] \) is a \( m \times n \) matrix and the Frobenius norm corresponds to the Euclidean norm when the matrix is considered as a vector.
The atomic norm is a type of norm used in fields such as optimisation and signal processing, where the atomic norm is generally designed to reflect the structural properties of a vector or matrix.
Structural Learning is a branch of machine learning that refers to methods for learning structures and relationships in data, usually in the framework of unsupervised or semi-supervised learning. Structural learning aims to identify and model patterns, relationships, or structures present in the data to reveal the hidden structure behind the data. Structural learning targets different types of data structures, such as graph structures, tree structures, and network structures.
This section discusses various applications and concrete implementations of structural learning.
Overlapping group regularization (Overlapping Group Lasso) is a type of regularization method used in machine learning and statistical modeling for feature selection and estimation of model coefficients. In this case, the feature is allowed to belong to more than one group at the same time. This section provides an overview of this overlapping group regularization and various implementations.
Labeling of image information can be achieved by various machine learning approaches, as described below. This time, we would like to consider the fusion of these machine learning approaches and the constraint satisfaction approach, which is a rule-based approach. These approaches can be extended to labeling text data using natural language processing, etc.
A topic model is a statistical model for automatically extracting topics (themes or categories) from large amounts of text data. Examples of text data here include news articles, blog posts, tweets, and customer reviews. The topic model is a principle that analyzes the pattern of word occurrences in the data to estimate the existence of topics and the relevance of each word to the topic.
This section provides an overview of this topic model and various implementations (topic extraction from documents, social media analysis, recommendations, topic extraction from image information, and topic extraction from music information), mainly using the python library.
DBSCAN is a popular clustering algorithm in data mining and machine learning that aims to discover clusters based on the spatial density of data points rather than assuming the shape of the clusters. This section provides an overview of this DBSCAN, its algorithm, various application examples, and a concrete implementation in python.
Multimodal search integrates multiple different information sources and data modalities (e.g., text, images, audio, etc.) to enable users to search for and retrieve information. This approach effectively combines information from multiple sources to provide more multifaceted and richer search results. This section provides an overview and implementation of this multimodal search, one using Elasticsearch and the other using machine learning techniques.
Elasticsearch is an open source distributed search engine for search, analysis, and data visualization that also integrates Machine Learning (ML) technology and can be leveraged for data-driven insights and predictions. It is a platform that can be used to achieve data-driven insights and predictions. This section describes various uses and specific implementations of machine learning technology in Elasticsearch.
Raspberry Pi is a Single Board Computer (SBC), a small computer developed by the Raspberry Pi Foundation in the UK. Its name comes from a dessert called “Raspberry Pi,” which is popular in the UK.
This section provides an overview of the Raspberry Pi and describes various applications and concrete implementation examples.
- Hello World of Neural Networks, Implementation of Handwriting Recognition with MNIST Data
As a hello world of deep learning technology, concrete implementation and evaluation of handwriting recognition technology for MNIST data by pyhton/Kera.
- Deep learning for computer vision with python and Keras (1) Convolution and pooling
In this article, we will discuss convolutional neural networks (CNNs), also known as cnvnet, a deep learning model that has been used almost without exception in computer vision applications. In this paper, we describe how to apply CNNs to the image classification problem of MNIST as handwritten character recognition.
- Deep learning for computer vision with python and Keras (2) Improving CNNs by Data Expansion with Small Amount of Data
We apply two more basic methods for applying deep learning to small data sets. One is feature extraction with pre-trained models, which improves the correctness rate from 90% to 96%. The second is fine tuning of the learned model, which will result in a final correctness rate of 97%. These three strategies (training a small model from scratch, feature extraction using the trained model, and fine tuning of the trained model) are some of the props that can be used when using a small dataset for attrition classification.
The dataset we will use is the Dogs vs Cats dataset, which is not packaged in Keras. This dataset will be the one provided by Kaggle’s Computer Vision Kompetition in late 2013. The original dataset can be downloaded from the Kaggle web page.
- Deep learning for computer vision with python and Keras (3) Improving CNNs using trained models.
In this article, we will discuss how to improve CNNs by using learned models. VGG16 is a simple CNN architecture widely used in ImageNet, which is a learned model consisting of classes representing animals and everyday objects. VGG16 is an older model, not quite up to the state of the art, and a bit heavier than many of the latest models.
There are two ways to use a trained network: feature extraction and fine-tuning.
- Deep learning for computer vision with python and Keras (4) Visualization of CNN training data
Since 2013, a wide range of methods have been developed to visualize and interpret these representations. In this article, we will focus on three of the most useful and easy-to-use methods.
(1) Visualization of the intermediate outputs of a CNN (activation of intermediate layers): This provides an understanding of how the input is transformed by the layers of the CNN and provides insight into the meaning of the individual filters of the CNN. (2) Visualization of CNN’s filters: To understand what kind of visual patterns and visual concepts are accepted by each filter of CNN. (3) Visualization of a heatmap of class activation in an image: This will allow us to understand which parts of an image belong to a particular class, and thus to localize objects in the image.
- Advanced Deep Learning with PyTorch(OpenPose, SSD, AnoGAN,Efficient GAN, DCGAN,Self-Attention, GAN, BERT, Transformer, GAN, PSPNet, 3DCNN, ECO)
Specific implementation and application of evolving deep learning techniques (OpenPose, SSD, AnoGAN, Efficient GAN, DCGAN, Self-Attention, GAN, BERT, Transformer, GAN, PSPNet, 3DCNN, ECO) using pyhtorch.
Actual Removal of Noise from Image Information Using Sparseland Model
Theory
Artificial Intelligence (AI) has great influence in the field of education and has the potential to transform teaching methods and learning processes. Below we discuss several important aspects of AI and education.
- Overview of Image Recognition(1)History and overview of image recognition technology
Image recognition is the art of understanding what is in an image. Image recognition has a wide range of applications, including character recognition, diagnostic support using medical images, detection from surveillance cameras, image and video search on the Internet, product inspection, personal information recognition from faces and fingerprints, sports image analysis, robot vision, automatic driving of automobiles, and human interface using motion recognition. A wide range of applications are possible. The performance of vision sensors for capturing images has been greatly improved in recent years, and they can input very rich information at a low cost.
In order to explain what image recognition is, we will briefly summarize the history of image recognition technology.
- Overview of Image Recognition(2)Overview of the steps in image recognition
The processing procedure of general class recognition is divided into two major modules: image feature extraction and classification. Image feature extraction is further divided into local feature sampling and description, statistical feature extraction, coding, and pooling. These procedures are connected and processed in series.
In this section, we will give an overview of each procedure.
- Local Features(1)Overview of local features and various filtering processes
The first part of the image recognition process is the extraction of local features that focus on local regions of the image and describe their contents. The process of extracting local features can be divided into detection in the first half and description in the second half. Detection is the process of capturing points in the image such as corners and edges, while description is the process of representing the local region around the points obtained in the detection process. The algorithm to find the points to focus on in the former is called a detector, and the vector described in the latter is called a descriptor.
A local feature is a feature that represents a small local region in an image, rather than the entire image. On the other hand, features that represent the entire image are called global features. In order to find a specific object in an image, comparison of local features is more effective than global features.
Local feature extraction consists of detection, which captures feature points in the image, and description, which represents the region around the feature points. The detection of feature points can be divided into two methods: one that captures points with characteristic shapes, such as corners and edges of objects (sparse sampling), and one that extracts feature points at regular intervals (dense sampling). Typical detectors include edge detectop, corner detector, and blob detector.
- Local Features(2)Detectors(Edge, corner and blob detectors)
An edge detector is a detector that captures points like the edges of an object. However, it requires a very advanced recognition function to determine whether the detected point is an edge of an object or not, so here we consider points with sudden changes in brightness as edges and extract them from the image.
The corner detector is a detector that finds the corner-like points of an object, and like the edge detector, it only detects the corner-like points of an object and does not actually judge whether it is a corner or not. The basic principle is that a corner is a point where the luminance changes significantly in two orthogonal directions.
Blob detectors focus on a certain small area and detect blobs, which are the areas where the situation in the surrounding area differs from that of the small area. For example, the luminance of a certain small area is high and that of the surrounding area is low, or the color of a certain small area is red and that of the surrounding area is blue.
- Local Features(3)Descriptors(SIFT,SURF,BRIEF,BRISK,HGO,GIST)
The process of converting the contents of a local region into information that is advantageous for recognition is called description, and the described information is called descriptor. A descriptor is generally represented as a vector v∈ℝD. Descriptors of local regions are called local descriptors. In order to obtain information that is advantageous for recognition, we extract the shape and texture information of the local region. A variety of methods have been proposed as descriptors.
Raw pixel descriptor is the simplest local descriptor, which is a vector of the local region information. The local binary pattern (LBP) is a descriptor that expresses the texture information of a local region, in which the difference in luminance between the center pixel and its surrounding pixels is calculated and assigned a binary pattern of 0 or 1 according to its code.
Descriptors using local luminance gradient histograms (SHIFT descriptors, HGO descriptors, etc.) Since luminance gradients have the effect of extracting edges, local luminance gradient histograms represent shape information and are robust to small rotations because the direction of the luminance gradient is quantized.
- Statistical Feature Extraction (PCA,LDA,PCS,CCA)
In actual images, some disturbance or noise is added, and if we use local features obtained from images that are affected by disturbance as they are, we may not be able to obtain the expected recognition accuracy. Therefore, statistical feature extraction is necessary to convert the observed data into features that are advantageous for recognition based on the established statistical structure of the data.
Statistical feature extraction means that the extracted local features are further extracted based on the probability statistical structure of the data, and transformed into robust features that are not easily affected by noise or disturbances. Statistical feature extraction can be applied not only to local features but also to various features in image recognition.
Statistical feature extraction can be classified according to the presence or absence of external criteria, i.e., teacher information, such as which class the data belongs to. When there is no external criterion, principal component analysis is used as a feature extraction method. When there is an external criterion, Fisher’s linear discriminant analysis is used for feature extraction in class recognition, the regular modified correlation distribution is used for bivariate correlation maximization, and the partial least squares method is used for bivariate covariance maximization. Although these seem to be different methods at first glance, they are deeply related to each other.
- Coding and Pooling (BoVW、GMM)
The operation of converting local features into vectors with a valid number of dimensions for recognition is called coding. The operation of combining multiple post-coding feature vectors existing in an image region into a single vector is called pooling.
Specific coding assumes that the data is sampled from a certain probability distribution, estimates the probability distribution, and extracts the coding function using the estimated distribution.
Pooling methods include average pooling, which calculates the average value of the target vector, and max pooling, which calculates the maximum value of each element of the vector.
There are two main advantages of pooling: first, even if the local features obtained from the image are different from each other in the figure below, the feature vector of the same dimension can be obtained by pooling; second, since the position information of the local features in the image region to be pooled is not taken into account, position-invariant features can be obtained. The second point is that it does not take into account the location information of the local features in the image region where pooling is performed, so it can obtain position-invariant features.
- Classification(1)Algorithm of the classifier(Bayes Decision Rule)
The input image becomes a feature vector after a series of processing. The final step of class recognition is classification, which assigns a class (e.g., “dog” or “cat”) to this feature vector. The algorithm that performs classification is called a classifier.
The algorithm for classification is called a classifier. In this section, we will discuss the Bayes decision rule for constructing a classifier.
- Classification(2)Optimization process(Gradient Descent Method, Newton’s Method, Perceptron, SVM)
Continuing from the previous article, we will discuss classifiers using the perceptron, deep learning, and SVM.
- Classification(3)probabilistic discriminant function(Logistic, Softmax Regression) and local learning(K-nearest neighbor method, kernel density estimation)
When considering class recognition, if we can predict the posterior probability of a class whose discriminant function takes a value between 0 and 1, we can quantify the degree to which the input data belongs to the target class. However, since the output of the discriminant function ranges from -∞ to +∞, it is difficult to directly interpret it as a posterior probability. Therefore, we can use a probabilistic discriminant function that extends the linear discriminant function to predict the posterior probability of a class. In this case, the probabilistic discriminant function is used. Logistic regression and softmax regression, which are approaches using probabilistic discriminant functions, are important elements of neural networks.
- Classification(4)collective learning(Ensemble Learning, Random Forest) and evaluation of learning results(Cross-validation method)
When the data is distributed in a complex way in the feature space, a nonlinear classifier becomes effective. To construct a nonlinear classifier, kernel methods and neural networks can be used. In this section, we describe ensemble learning, which constructs a nonlinear classifier by combining multiple simple classifiers. Collective learning is also called ensemble learning.
As collective learning, we describe bagging, which generates subsets from the training data set and trains a predictor on each subset. This method is particularly effective for unstable learning algorithms. An unstable learning algorithm is one in which small changes in the training data set have a large impact on the structure and parameters of the predictor being learned. Neural networks and decision trees are examples of unstable learning algorithms.
The bootstrap method is a method of generating diverse subsets from a finite set of data. The bootstrap method is a method to generate M new data sets by repeating random recovery extraction from the data set M times.
- Convolutional neural networks(1)Forward and back propagation algorithms and mini-batch
Local feature extraction, statistical feature extraction, coding and pooling are each considered as one module, and the structure in which these modules are stacked on top of each other in multiple levels is called a deep structure. The method of learning this deep structure from input to output in an end-to-end manner is called deep learning. In deep learning, it is common to design the constituent modules by using neural networks, and it is common to design the deep structure using neural networks by using deep neural networks, and it is common to design the deep structure using neural networks by using deep neural networks. Deep neural networks are referred to as deep neural networks. By using deep learning, it is possible to build a system that predicts the desired output for input data, even without being familiar with the local feature extraction and coding methods mentioned above.
In this article, we will discuss forward and back propagation algorithms and mini-batch as an overview of deep learning techniques.
- Convolutional neural network(2)Overview and implementation of CNN
Continuing from the previous article, we will discuss the theoretical overview and implementation of convolutional neural networks (CNNs), which are frequently used for image recognition in deep learning.
- Object detection Sliding Window Method and Negative Example Sequential Selection with Exampler-SVM, R-CNN
Object detection aims to find a rectangular region in an image that surrounds an object such as a person or a car. Many object detection methods propose multiple candidate object regions and use object class recognition methods to determine which object these regions are classified as. Since the number of candidate object regions proposed from images is often huge, methods with low computational cost are often used for object class recognition.
Sliding window method, selective search method, and branch-and-bound method are the methods to propose object region candidates from images. There are also several methods to classify them, such as Exampler-SVM, Random Forest, and R-CNN (regious with CNN feature).
- Instance recognition and retrieval(1)Instance search using BoVW
While class recognition involves predicting the class to which a target object belongs, instance recognition is the task of identifying the target object itself. The central task of instance recognition is the image retrieval problem, which is to quickly find an image in a database from an input image. Instance recognition is the task of identifying the object itself, such that when we see the Tokyo Tower, we do not recognize it as a radio tower, but as the Tokyo Tower. This can be achieved by searching the database for images that show the same object as the one in the input image.
The implementation of instance recognition is as follows: 1) extract local features from a set of stored images and create an image database, 2) extract local features of the query image, 3) take one local feature of the query image and compare it with all local features in the image database. Cast one vote for the image in the database that has the most similar local features. The object in the image with the most votes in the database is recognized as the object in the query image.
- Instance recognition and retrieval (2) General image retrieval
The problem of finding images in the database that are similar to the image represented by the feature vector x is called similar image search or image retrieval, and is one of the central problems in instance recognition.
The simplest way to achieve image retrieval is to rank the images in the database by measuring the distance between the query image and all the images in the database and sorting them in ascending order. However, when the number of images in the database becomes huge, this method becomes impractical because it takes too much computation time. In this paper, we will discuss efficient search methods using tree structure, binary code conversion, and direct product quantization.
An overview of dictionary generation by machine learning based on sparse land models, rather than dictionaries derived from signal processing knowledge (which have a DCT basis in JPEG), for sparse representation of images using dictionary data.
In this article, we will discuss sparse regularization with duplicates. Sparse regularization with duplicates is a cloud combination of sparse regularization terms, e.g., with respect to subvectors or linear transformations of a vector Ω ∈ ℝd, and has applications in image processing, statistics, tensor decomposition, and others.
The autoencoder is trained by giving the same vector to the input and output layers. The idea is to make the number of neurons in the middle layer smaller than in the input and output layers, so that the output of the middle layer can be extracted and compressed as a feature of the data.
Here is an example of the application to handwritten character recognition, along with Hinton’s paper.
Introduction to nearest neighbor methods, decision trees, and neural networks as basic algorithms for pattern recognition.
Introduction to data compression algorithms used in image information (JPEG), etc.
On Edge Extraction Techniques Using Variational Methods from AAAI Classic Proc.
Linear dimensionality reduction (linear dimensionality reduction) is a basic technique for reducing the amount of data, extracting feature patterns, and summarizing and visualizing data by mapping multidimensional data to a low-dimensional space. In fact, it is known empirically that, for many real data, a space of dimension M, which is much smaller than the dimension D of the observed data, is sufficient to represent the main trends of the data, so the idea of dimensionality reduction has been developed and utilized in various application fields, not limited to machine learning.
The methods described here are closely related to techniques called probabilistic principal component analysis, factor analysis, or probabilistic matrix factorization. Although closely related to techniques such as probabilistic principal component analysis, factor analysis, or probabilistic matrix factorization, we will focus here on simpler models that are simpler than commonly used methods.
In addition, as a specific application here, we will also conduct simple experiments on image data compression and interpolation of missing values using the linear dimensionality reduction model. The ideas of dimensionality reduction and missing value interpolation are common to models such as nonnegative matrix factorization and tensor decomposition.
We discuss two mathematical properties of the atomic norm and the equivalence of the norm with the convex hull of the atom set as the unit sphere and the representation of the dual norm of the atomic norm. Although the atomic norm is mathematically sophisticated and contains norms that induce various sparsity properties, it is difficult to compute the norm itself or the prox operator on the norm, except in special cases such as the L1 norm, group L1 norm, and trace norm. We will discuss the Frank-Wolfe method, which is effective when optimization with a certain degree of accuracy is sufficient, and the dual alternating direction multiplier method, which is effective when a solution with a slightly higher accuracy is desired. Finally, a concrete example of foreground image extraction using robust principal component analysis is presented.
- Overview of Multi-Task Learning and Examples of Applications and Implementations
Multi-Task Learning is a machine learning method that simultaneously learns multiple related tasks. Usually, each task has a different data set and objective function, but Multi-Task Learning aims to incorporate these tasks into a model at the same time so that they can complement each other by utilizing their mutual relevance and shared information.
Here, we provide an overview of methods such as shared parameter models, model distillation, transfer learning, and multi-objective optimization for this multitasking, and discuss examples of applications in natural language processing, image recognition, speech recognition, and medical diagnosis, as well as a simple implementation in python.
コメント