Automatic Cast Shadow Detection

Abstract

For human eyes it is not difficult to distinguish shadows from objects. However in most situations, identifying shadows by computer is a challenging research problem which is a type of segmentation problem in digital image processing. In real time systems, the automated path inding robots can easily detect obstacle and take alternate path but the problem is that they also consider shadow as a part of obstruction. In this project a hybrid algorithm is proposed to detect and extract cast shadow from a still image using fuzzy divergence thresholding and C1C2C3 color invariant feature. The proposed algorithm consists of two modules. The first module gives the edge map of shadow whereas the second module gives the object edge map. Exploiting the geometric property of shadow and these two edge maps, cast shadow region is extracted. Finally it is remapped on original image to visualize the extracted region. Some of the important applications can be in processing satellite images, object tracking in video and, image matching and enhancement.