Search This Site

Image recognition via codebooks and contour reinforcement

Sumeet Gupta, Vishesh Gaurav and Ramani, S.

This paper presents image recognition techniques which is widely studied by researchers with various methods such as the bag-of-visual-words model, Contour Reinforcement and LRSC. These features of images are followed by classifier training and evaluation. This paper presents two such methods for image recognition. When using codebooks technique for image recognition, local features are randomly selected to learn the codebook accordingly. We take input from user and classify the input on the basis of stored codewords in our codebooks. The test results demonstrate that by taking general and class specific codebooks with the images, we are ready to display the distinctions among various fine-grained classes. In our other method which is sketch-based image retrieval (SBIR) the most challenging part was to identify the similarity between a sketch and a picture. In order to overcome this drawback, we propose SBIR based salient contour reinforcement. We divide the image contour into two different types the global contour map and the salient contour map which is useful for searching the objects in pictures. Additionally, to support the two contour map a replacement descriptor is proposed specifically angular radial orientation partitioning (AROP) feature. It utilizes orientation info of the pixels of edges into contour maps to spot the spatial relationships. This AROP feature is economical and effective way to get false matches of native options between sketch and pictures, and may improve the retrieval performance.

Download PDF: