The effect of image enhancement on biomedical pattern recognition

Abstract

Image enhancement has been an area of active research for decades. Most studies were aimed at improving the quality of image display for better visualization. Yet few studies have been conducted to investigate the impact of image enhancement on biomedical pattern recognition. In this paper, we examine quantitatively the effect of image enhancement on the performance of biomedical pattern recognition. We apply the wavelet-based image enhancement technique developed in our earlier work [Y. Wang et al., Proc. ICASSP 2001, Salt Lake City, May, 2001], to a well-known biomedical pattern recognition problem: chromosome classification. Experiments were conducted on a test set of chromosome images before and after the enhancement, using the same feature measurement and classifier methods. The test results show that our image enhancement method substantially reduces the error rate of chromosome classification. We learn from this study that proper image enhancement leads to significantly improved recognition accuracy, and the quantification of performance improvement may be used as an objective measure of success for evaluating various image enhancement techniques.