Research On Shadow Detection And Removal Approach For Still Images

As a special form of image degradation, shadow can lead to loss or disruption ofmachine vision image information, and then results in unstable situations or evenfailure of edge extraction, object recognition and image matching in image processing.In this case, the research of shadow detection and removal get a lot of attention.The thesis concentrates on the research of shadow detection and removalmethods of the static images， and focus on the methods based on shadowcharacteristics. Firstly, we summarize the background and significance of the staticshadow processing algorithms of static images, summarize the development historyand status of different shadow detection and removal algorithms. Then we analyze thecauses, characteristics and spectral properties of shadow. Finally, we summarize thework of existing shadow detection and removal algorithms based on these theories.An improved shadow detection algorithm is presented in this paper by usingclustering method to merge divided areas and decrease over-segmentation. Besidesthat, based on the features of paired-regions, support vector machine(SVM) isestablished to classify them, and then the algorithm’s efficiency and shadow detectioneffect are all improved. And we use a method which is based on ratio of direct lightintensity and ambient light intensity to realize shadow removal. Simulation resultsindicate that the running time of the proposed algorithm is remarkably reducedcompared with the traditional one. Besides that, the shadow detection effect is moreaccurate for complicated texture images and the results of shadow removal isefficient.Finally, we present a shadow detection and removal method based on the ratio ofcolor, this method divide the whole image into the penumbra shadow and the umbraarea based on the method of matting. Then the shadow scale factor of umbra area andpenumbra shadow are calculated to realize shadow removal. The results of experimentdemonstrate the effectiveness of this algorithm.