Segmentation Using Saliency–color Mapping Technique

The accuracy of face image classification/recognition is absolutely based on the extraction of object of interest in the image. This can be achieved by identifying salient object and eliminating background and other unwanted details. Therefore, this paper proposes a method for object identification and segmentation, which is based on finding fused saliency maps by combining pixel-intensities, visual attracted location of the image and color in the CIEL*a*b* color space. This method uses quadrants and cluster center window of the image without transforming other domain like frequency domain, to obtain saliency maps, in order to minimize time, cost. The object of interest from the original image is segmented using spatial color mapping technique to map saliency image with actual input image using threshold. In order to get perfect segmentation, the input image is subjected to filter and enhanced. This method is effectively applied on data sources such as MSRA, IMM database, and random samplings. The obtained saliency map, subject to binarised, is compared with ground truth using statistical measures such as accuracy, precision and recall, and compared with other methods. As results of this comparison, the proposed method outperforms.