Gradient weighted CO-HOG for analysis of caudal vein structural changes in toxin exposed zebrafish embryo

In this paper, we propose an approach to detect abnormalities in the vascular development of zebrafish embryos due to exposure to toxins. Morphological changes due to toxin exposure is modeled based on the proposed gradient weighted co-occurrence histogram of oriented gradients (gCo-HOG). These features are compared to more commonly used histogram of oriented gradients (HOG) features and Co-HOG features that utilizes spatial distribution of neighboring pixels to capture spatial structure. The features are used to train a linear SVM classifier to identify morphological changes in a dataset of zebrafish embryo images. Results show that proposed gCo-HOG based technique clearly outperforms HOG and Co-HOG features. The importance of descriptor lies in the fact that features are discriminative enough to capture subtle morphological changes between normal anatomy and toxin induced changes.