|
Image retrieval has wide range of applications in various domains such as medical diagnosis. Retrieval based on content of the image eliminates the laborious manual task of image annotation needed otherwise. Content based image retrieval (CBIR) permits the retrieval of images of similar content or information. It has become an active topic of research since for the last decade. One of the major issues in CBIR is the difficulty in the representation of the meaning or semantics of the scenes. CBIR technology that operates on the basis of low level image semantics cannot be directly related to the descriptive semantics that is used by human for deciding image similarities. The lowlevel semantic of the image consists of texture, colour, intensity and shape of the object inside an image. However, only one type of feature extraction results in poor performance. There is substantial increase in retrieval accuracy when combinations of these techniques are used in an effective way. In this paper, we propose an improvement in CBIR technology using different feature extraction methods; two features based on colour and another two feature computed by applying the texture feature using Gabor wavelet and Discrete Cosine Transform coefficients of the image. For similarity matching between the images, Manhattan distance (City Block) is used. The experimental results on WANG database showed higher retrieval efficiency (in terms of precision) when compared with existing methods using texture and colour features.
|
