Hybrid Approach of Medical Image Classification Using Sparsity Regulatization and BAT Algorithm

Abstract

Classification has gained popularity in research community due to its wide variety of applications in different fields. The complexity in extraction of information from images and use of this information for classification task has made image classification a tedious job.  The Proposed method of classification uses SIFT algorithm for extracting rotation and scale invariant features. Using the concept of Deep Neural Network sparse coding technique is used for generating codebooks from the extracted features. Sparse codes give an intermediate representation between local codes and dense codes. These codes have capability to extract information at different levels and with varying amount according to the type of input. Instead of using L1 norm another popular regularizer is used in this paper, which maintains group sparsity for non-overlapping groups. Pooling operation is applied on the sparse coded features. Bat algorithm is used on these pooled features for classification of medical images. Experimental results prove the fruitfulness of our proposed method in medical image classification.