A multi-channel based illumination compensation mechanism for brightness invariant image retrieval

Abstract

The image retrieval is still challenging to retrieve the most similar images of a given image from a huge database more accurately and robustly. It becomes more challenging for the images having drastic illumination differences. Most of feature descriptor having better retrieval performance degrades in the case of illumination change. To circumvent this problem, we compensated the varying illumination in the image using multi-channel information. We used Red, Green, Blue channel of RGB color space and Intensity channel of HSI color space to remove the intensity change in the image. Finally, we designed an illumination compensated color space to compute the feature descriptor over it. The proposed idea is generic and can be implemented with the most of the feature descriptor. We used some state-of-the-art feature descriptor to show the effectiveness and robustness of proposed color transformation towards uniform and non-uniform illumination change. The experimental results suggest that proposed brightness invariant color transformation can be applied effectively in the retrieval task.