Investigating Image Enhancement in Pseudo-Foreign Fiber Detection

The detection of pseudo-foreign fibers in cotton based on AVI(Automatic Visual Inspection) is crucial to improve the accuracy of statistics and classification of foreign fibers. To meet the requirement of textile factories, a new platform is introduced in which cotton bulks are floating with relative high speed of six meters per second, and the throughput of detected lint could be above 20kg per hour. However, images captured by the new platform are blurred and not clear enough for post processes such as segmentation, feature extraction, target identification and statistics. Because thickness of the moving cotton bulks are not uniform, a part of or the whole object of pseudo-foreign fibers are blocked. Thus image enhancement algorithms should be investigated and implemented. In this paper the characteristics of the images acquired by the new platform are analyzed, and several image enhance algorithms are studied and compared on effectiveness and efficiency, which include Histogram Equalization, Wavelet Based Normalization, Homomorphic Filtering, Single Scale Retinex(SSR), Multiscale Retinex(MSR) and Variational Retinex. Result indicated that the Variational Retinex has a better performance and should be implemented in on-line pseudo-foreign fibers detection.