Evaluation For Robustness Of Shadow Featuresand Shadow Detection Algorithm

Shadows in images bring some undesirable problems in many computer vision tasks, such as image segmentation, object recognition, edge detection. Shadow detection can avoid above mentioned problems and can aid shadow removal. Therefore, shadow detection is a hot topic in both image processing and computer vision. Compared with detection moving shadows, detecting static shadows is a more challenging task.In this paper, we divide recent static shadow detection approaches into three categories: model based methods, intrinsic image based methods and statistical learning based methods. We first survey and summarize the current status in these researches, then give detailed description in the following chapters. At last, we further discuss their open problems and future development.The main innovation of this paper is as follows:1、Currently, several review articles about moving shadows have been published. However, a suvey paper is not found for static shadow detection algorithms. This is the first time to classify and summarize research status of static shadows, as well as the current problems and trends.2、For the current status of imperfect shadow data sets, providing a massive shadow data sets and shadow position by manually tagging for researchers to use. Several common color spaces are compared, analyzed their insensitivity to light, shadows sensitivity and spatial differences in the internal channel are analyzed. The selection of the color space for shadow processing are given.3、In recent years, mainly static shadow detection using multiple features, and then use the classifier after learning to detect shadows. But we can not know which feature is the good one, and who play a major role. Therefore, we propose a shadow feature evaluation method, commonly used to compare characteristics of the shadow, identified to play a major role in the detection result of good features.4、According to Chapter IV find good shadow feature, we propose a new shadow detection algorithm which can effectively detect shadows.