Model-Based Approach for Shadow Detection of Static Images

Abstract

Listen

Translate article

The existing computer vision systems accept computer-based images for processing. These images can be processed differently depending upon the types of applications. One such application that maps to the domain of our work is tracking in which the image has a number of objects of which only the object of interest is mapped for tracking. Due to the light sources and existence of surrounding light conditions, the object of interest, if opaque, creates a hard cast shadow on the background while if the object is semi-opaque, it creates soft cast shadows on the background. These shadows resemble very closely to the shape of the object. Thus, there are chances that they may be mistaken for the object leading to the problem of false tracking. To avoid this problem, the shadows have to be detected in the computer vision system before they enter the processing unit. To add to this, the computer vision systems cannot differentiate between an object and its shadow and thus passes both of them. So it was required as a necessary and important stage to incorporate the shadow detection and elimination stage between the computer vision and processing stage. This stage which is a complete system on its own has its own input stage, processing, output and evaluation stage. The scope of the work presented in the paper was for two methods, i.e. colour based and texture based along with justified evaluation of the results obtained by them.