Color Histograms Adapted to Query-Target Images for Object Recognition across Illumination Changes
- PDF / 1,780,434 Bytes
- 9 Pages / 600 x 792 pts Page_size
- 34 Downloads / 169 Views
Color Histograms Adapted to Query-Target Images for Object Recognition across Illumination Changes Damien Muselet Laboratoire LAGIS, UMR CNRS 8146, Universit´e des Sciences et Technologies de Lille, Cit´e Scientifique, Bˆatiment P2, 59655 Villeneuve d’Ascq, France Email: [email protected]
Ludovic Macaire Laboratoire LAGIS, UMR CNRS 8146, Universit´e des Sciences et Technologies de Lille, Cit´e Scientifique, Bˆatiment P2, 59655 Villeneuve d’Ascq, France Email: [email protected]
´ Jack-Gerard Postaire Laboratoire LAGIS, UMR CNRS 8146, Universit´e des Sciences et Technologies de Lille, Cit´e Scientifique, Bˆatiment P2, 59655 Villeneuve d’Ascq, France Email: [email protected] Received 14 January 2004; Revised 24 November 2004 Most object recognition schemes fail in case of illumination changes between the color image acquisitions. One of the most widely used solutions to cope with this problem is to compare the images by means of the intersection between invariant color histograms. The main originality of our approach is to cope with the problem of illumination changes by analyzing each pair of query and target images constructed during the retrieval, instead of considering each image of the database independently from each other. In this paper, we propose a new approach which determines color histograms adapted to each pair of images. These adapted color histograms are obtained so that their intersection is higher when the two images are similar than when they are different. The adapted color histograms processing is based on an original model of illumination changes based on rank measures of the pixels within the color component images. Keywords and phrases: color, object recognition, illumination, adapted color histograms, rank measures.
1.
INTRODUCTION
1.1. Object recognition by color histogram analysis Object searching in a database of color images, which is aparticular problem of color image retrieval, is identical to appearance-based object recognition. In this framework, the recognition problem can be stated in terms of finding among all the target images of a database, those which contain the same object as that represented by the query image. Each of these images contains one single object placed on a uniform background. In this context, the image indexing scheme consists in extracting robust and efficient characteristic indices from the target and query images. These indices are typically derived from the shape [1], the texture [2], or the color properties [3] of the objects. Object recognition is performed by means of a matching scheme which compares the indices of the query image with those of the target images. The matching
scheme is based on a similarity measure between these indices. The target images are ranked with respect to their similarity measures with the query image, in order to determine those which contain the same object as that represented by the query image. One of the most widely used image indices based on the color distribution is the color histogram [
Data Loading...
