Image Classification: Step-by-step Classifying Images with Python and Techniques of Computer Vision and Machine Learning

* Research Fields: Computer Vision and Machine Learning.
* Book Topic: Image classification from an image database.
* Classification Algorithms: (1) Tiny Images Representation + Classifiers; (2) HOG (Histogram of Oriented Gradients) Features Representation + Classifiers; (3) Bag of SIFT (Scale Invariant Feature Transform) Features Representation + Classifiers; (4) Training a CNN (Convolutional Neural Network) from scratch; (5) Fine Tuning a Pre-Trained Deep Network (AlexNet); (6) Pre-Trained Deep Network (AlexNet) Features Representation + Classifiers.
* Classifiers: k-Nearest Neighbors (KNN) and Support Vector Machines (SVM).
* Programming Language: Step-by-step implementation with Python in Jupyter Notebook.
* Processing Units to Execute the Codes: CPU and GPU (on Google Colaboratory).
* Major Steps: For algorithms with classifiers, first processing the images to get the images representations, then training the classifiers with training data, and last testing the classifiers with testing data to get the prediction accuracies; for algorithms with networks, first building a network, then training the network with training data, and last testing the network with testing data to get the prediction accuracies.
* Main Results: For the testing data, the prediction accuracies vary between about 30% and 90%, while the time consumptions varied from several seconds to more than one hour. Considering both of the criteria, the Pre-Trained AlexNet Features Representation plus a Classifier was concluded as the best algorithm.
* Detailed Description:
This book implemented six different algorithms to classify images with the prediction accuracy of the testing data as the primary criterion (the higher the better) and the time consumption as the secondary one (the shorter the better). The accuracies varied between about 30% and 90%, while the time consumptions varied from several seconds to more than one hour. Considering both of the criteria, the Pre-Trained AlexNet Features Representation plus a Classifier, such as the k-Nearest Neighbors (KNN) and the Support Vector Machines (SVM), was concluded as the best algorithm.
The six algorithms are: Tiny Images Representation + Classifiers; HOG (Histogram of Oriented Gradients) Features Representation + Classifiers; Bag of SIFT (Scale Invariant Feature Transform) Features Representation + Classifiers; Training a CNN (Convolutional Neural Network) from scratch; Fine Tuning a Pre-Trained Deep Network (AlexNet); and Pre-Trained Deep Network (AlexNet) Features Representation + Classifiers.
The codes were written with Python in Jupyter Notebook, and they could be executed on both CPUs and GPUs.