PROJECT TITLE :

Regularization on Augmented Data to Diversify Sparse Representation for Robust Image Classification

ABSTRACT:

The process of image classification is an essential part of today's computer vision systems. Due to the robustness of the sparse representation-based classification method, it has garnered a lot of attention in recent years. However, regularization and data augmentation are both powerful techniques, but they are currently being used in isolation when it comes to the optimization of sparse learning systems. We have reason to believe that the combination of data augmentation and regularization can bring about a significant improvement in the accuracy of image classification. Regularization on Augmented Data (READ) is a novel framework that we propose in this article. It generates diversification in the data by making use of generic augmentation techniques in order to implement robust sparse representation-based image classification. When the training data are augmented, READ applies a separate regularizer, in this case either l1 or l2, to the augmented training data in addition to the original data. This ensures that regularization and data augmentation are both utilized and improved at the same time. We present a comprehensive theoretical analysis on how to optimize the sparse representation by both l1 -norm and l2 -norm with the generic data augmentation, and we demonstrate the effectiveness of our approach through a number of in-depth experiments. When it comes to the use of deep features, the results that READ achieved on a variety of facial and object datasets demonstrate that it outperforms many other state-of-the-art methods.