An Effective Content-Based Image Retrieval Using Color, Texture and Shape Feature

Content based image retrieval is an active area of research for more than a decade. In this paper, we have proposed an effective way of extracting color, texture, and shape features from image and combine them in a way that ensures higher retrieval efficiency. For extraction of color features, images are divided into non-overlapping blocks, and dominant color of each block is determined using k-means algorithm. For extracting gray-level co-occurrence matrix (GLCM)-based texture features, each pixel in the image is replaced by average value of its neighborhood pixels. These average values are further quantized into 16 levels, for better and efficient representation of texture in the database. Finally, Fourier descriptors are extracted from the segmented image and are used to represent the shape of objects, as they have better representation capability and robust to noise, than other shape descriptors. The feature vector formed by combining all these is used to represent image in the database. We have tested our approach on wang dataset. Experimental results show that the present scheme has achieved higher retrieval accuracy on representative color image databases.