In this paper, we proposed the improved results of the face retrieval process by normalizing the local features of the face from the input image and using the clustering of features extracted from the overall unique shape of the face.To improve the accuracy of face retrieval, facial details were extracted by normalizing local features. In order to normalize the regional features, the feature refinement process considering the facial features was used in the GIST-based feature extraction results. After that, face shape estimation based on the eigenvector values was performed to extract the overall unique shape of the face, and it was used as an additional weight by clustering the shape estimate values of each image by performing it on the entire database image.In the case of using a conventional image retrieval method for face retrieval, there is a problem that the accuracy is relatively low due to a simple comparison method of detail information such as color, border, and texture without considering the characteristics of the face. To improve this, algorithms such as SIFT and SURF using shape features have been proposed, but the results are inaccurate due to the relatively low number or accuracy of finding feature parts from similar images. The proposed method showed more improved to face search results by estimating similar images using the weights considering the refinement of local features and the overall unique shape of the face.By using the proposed method, we improved the result of searching for the same face by 6% compared to the number of search images.
Volume 11 | 06-Special Issue
Pages: 2035-2042