Content based image retrieval with combined color and texture features

Abstract

The number of digital image and video databases in the Internet and other information sources are growing rapidly. Indexing these huge database by name is a very laborious job. To serve this purpose the concept of Content Based Image Retrieval (CBIR) has been introduced which uses visual contents to search images from large scale image databases according to users' interests. We proposed a new image content descriptor that considers both texture and color information of an image and works on each color plane of a RGB image at the same time to reduce the feature vector size. In a n n neighborhood of an image we nd the dissimilarity of the center pixel with all other neighboring pixels using the Euclidean Dissimilarly method and Cosine similarity method. Then based on a threshold value we set some binary values to the dissimilarity values and nally get the feature vector. In this process all the three color planes (R, G and B) are considered. That's why our feature descriptor contains both color and texture information. Using this feature descriptor we train the system. When a query image comes we measure the distance between the query image and the database images using the Chi-square method and determine the most similar images of the query image. The experiments show that our method gives better image retrieval rate while using a small feature vector which is computationally very e cient.