A novel approach for searching of a color facial image based on the similarity of complexion
Sudipto Halder, Mita Nasipuri, Abul Hasnat, Debotosh Bhattacharjee · 2013
This paper proposes a novel methodology to find a facial image similar in complexion of a given query image from a facial image database. There are a lot of methodologies described in literature to find the distance between two images. But none of the existing methodologies has focused on the complexion based searching and that is why they don't give satisfactory result in this aspect. But the proposed methodology, described in this paper, shows a significant improvement in skin color based facial image searching result. To validate the result, obtained from the proposed methodology, an opinion poll of 100 peoples has been taken to decide the similar face of a given query facial image in terms of complexion based on visual effect. For each query image the poll has been conducted between the four facial images obtained by the Manhattan, Euclidean, Modified Euclidean Distance and the proposed methodology and one face between them has been selected using maximum voting system. In all the cases, the proposed methodology complies with the result of the opinion poll. The proposed system first extracts the skin area from a facial image by removing the background region and hair region and applies the proposed methodology on skin area of the facial images. The proposed methodology has a huge application in the area of archaeology, entertainment, law enforcement etc. and the method is especially useful for criminal investigation department during the searching of a query image from a set of images containing the faces of different race. The proposed method has been tested on 75 male and 25 female color facial images of standard FRAV2D database and has been implemented using Matlab 7.