Impact of Shadow Detection and Removal on Object Recognition Using Machine Learning from Images

Object recognition is an important research subfield in the area of Image analysis. Any algorithm to perform object recognition with near-human accuracy is still a challenging task especially from real life images captured by general purpose camera. i.e; images which are not captured in a very constrained environment as far as lighting and/or pose are concerned. Such images are bound to have some shadow component. This shadow component is to be correctly detected and removed by an image analysis algorithm so that they can be analyzed for the target application. Failure in doing so will lead to many problems. For e.g; It may result in wrong segmentation leading to increase in number of segments obtained. These segments may, in turn,, be treated wrongly as objects in the image and will, in turn, have an incorrect computation of features. So the shadow detection and removal is an inevitable pre-processing step for object recognition. Many approaches have been proposed in the literature for shadow detection and removal. The main contribution of this paper is twofold; first, it performs comparative study of three shadow detection and removal approaches. And second, manifests empirically, the need of shadow detection and removal process, as one of the pre-processing step, on the success and accuracy of object recognition algorithm. For the purpose of experiment fruit images are considered and they are recognized. Results are presented and conclusions are drawn.