Iterative local color normalization using fuzzy image clustering

The goal of this paper is to introduce a new fuzzy local iterative algorithm that matches local color statistics of a reference image to the distribution of the input image. Reference images are considered to have a desirable color distribution for a specific application. The proposed algorithm consists of three stages: (1) images clustering by fuzzy cmeans, (2) clusters' matching, and (3) color distribution transfer between the matching clusters. First, a color similarity measurement is used to segment image regions in the reference and input images. Second, we match the most similar clusters in order to avoid the appearing of undesirable artifacts due to differences in the color dynamic range. Third, the color characteristics of the reference clusters are transferred to the equivalent clusters in the input image by applying an iterative process. The new image normalization tool has several advantages: it is computationally efficient and it has the potential of increasing substantially the accuracy of segmentation and classification systems based on analysis of color features. Computer simulations indicate that the iterative and gradual color matching procedure is able to standardize the appearance of color images according to a desirable color distribution and reduce the amount of artifacts appearing in the resulting image.