On illumination invariant image indexing from compressed domain using XYX-LUV transform

Abstract

In our previous work, illumination invariant object recognition has been achieved efficiently and effectively by normalizing the three color bands. We further employed the compressed histogram of the chromaticity to arrive at a valuable representation of the object which can facilitate high recognition precision. The shortcoming of this method lies in the usage of uniform quantization scheme in obtaining the chromaticity, which is not in agreement with the human vision system. In this paper, we endeavor to use the XYZ-LUV transform to circumvent this problem. Furthermore, instead of using the uncompressed image to achieve the illumination invariant recognition, we conduct our coding directly in the DCT domain by using several coefficients for each macroblock. Desirable results have been observed through our preliminary experiments.