Research On The Key Technology Of Image Shadow Detection Based On Mean Shift Segmentation

Image shadow detection remains an extremely challenging problem on computerimage processing. When the image forming apparatus for image acquisition, the shadowwill be entered into the image scene, causing a decline in image quality and will impact oncomputer image processing. For example, shadow will reduce object segmentation, targettracking, target recognition result of the processing such as unstable or even fail in thevideo image, even degrading the aesthetic of the image. Otherwise shadow not onlyinterfere with the subsequent result of computer image processing but also increase thedifficulty of scene analysis. Shadow detection and removal is a effective solution todramatically degrade the effect of the shadow to improve the accuracy of the computerimage processing. Among them, the shadow detection is the premise of the shadowremoval. We research on the image shadow detection to provide the basis for shadowremoval.Salvador, who divided shadow detection algorithm into model-based shadow detectionand feature-based shadow detection. Feature-based shadow detection algorithm hadbecome the mainstream of the image shadow detection algorithm in a very wide range ofapplications due to less computation and real-time higher concern. Feature-based shadowdetection algorithm usually use a segmentation algorithm to split shadow image intoshadow and non-shadow, and then shadow detected by extracting effective feature. MeanShift segmentation algorithm as the main of the segmentation algorithm split shadow imageoften causing over-segmentation problem, so it will increase the difficulty of featureextraction. Obviously the performance of segmentation directly affect the shadow detection.On the features of treatment, the support vector machines SVM features classification andprediction had become the mainstream of the shadow detection algorithm. The shadowdetection model trained by features is used to detect shadow but all features are integratedinto a single model which was not conducive to an effective description of thecharacteristics of the shadow. In our segmentation algorithm, we combine OTSU andMean Shift segmentation to improve shadow segmentation algorithm. In order to describethe shadow characteristics by relevant regional features after the Mean Shift segmentation,we choose some useful shadow features and propose the SVM dual-model shadowdetection mechanism in this paper. Experimental results show that the proposedsegmentation algorithm and the SVM dual-model-based shadow detection have achievedbetter detection results in a variety of image scene.