Colour and Positional Quantisation for Content-Based Image Retrieval

Colour-based retrieval is one of the major content-based retrieval techniques. The main idea used for colour based retrieval techniques is to store the image in terms of a colour map which has a nite set of colour bins. An eecient retrieval scheme should preserve all the distinct colours of an image with a small number of colour bins. Uniform Quantisation is one of the major quantisation techniques used for content based retrieval. Uniform quantisation results in a very small number of attributes for the feature vector ; therefore, it allows the use of simpliied indexing structures. This makes the retrieval process very easy. One of the drawbacks of uniform quantisation is that it retrieves a large number of false positives. We propose an image retrieval technique based on uniform quantisation which has fewer false positives. The quantised images are stored in a reduced form and the positional information of the colour pixels is preserved and used in the retrieval process. We examine the eeciency of this method and compare it with the uniform quantisation method. The results show that the modiied technique performs better than uniform quantisation as it has a fewer number of false positives. This approach also reduces the space requirements for image indexing as we handle a smaller set of image pixels.