Most applications of computer vision such as image correspondence, image retrieval, object recognition, texture recognition, face and facial expression recognition, 3D reconstruction, etc. required matching of two images. In order to match two images, or in other words to find out the similarity/dissimilarity between two images, some description of image is required because the matching of raw intensity values of two images will be more time consuming and it will be affected by any small variations in its inherent properties such as brightness, orientation, scaling, etc. Thus, the images can be matched with its description derived from the basic properties of the image such as color, texture, shape, etc. This description is called as the feature descriptor/signature of the image. The main objectives of any descriptor are 1) to capture the discriminative information of the image, 2) to provide the invariance towards the geometric and photometric changes, and 3) to reduce the dimension of the feature to be matched.

However, in recent years, the Deep Learning methods such as Convolutional Neural Networks (CNN) have shown very promising performance over many computer vision problems such as image classification, object detection, image segmentation, face recognition, etc. Basically, CNNs learn the image features automatically from data in a hierarchical way. Thus, I have also started working over CNN for different problems of computer vision. I have designed the different CNN models as well as CNN layers for deep learning.