New solutions for face photo retrieval based on sketches

Abstract

The problem of face photo retrieval using sketches constructed based on a description provided by a witness is discussed. The status of this problem from primary concepts and the used terminology, to modern technologies for constructing sketches, real scenarios and search results is reviewed. The development history of systems for constructing facial composites (identikits and sketches) and the ideas realized in these systems are provided. The task of automatically searching through a database of original photo images using a face sketch is discussed, and the reasons of low performance of such search in real-world scenarios are brought to light. Requirements to databases of sketches in addition to the existing benchmark face databases and also methods of creation of such databases are formulated. Within this framework the methods for generation a population of sketches from the initial sketch to improve the performance of sketch-based photo image retrieval systems are discussed. A method to increase the index of similarity in pairs sketch-photo based on computation of an average sketch from the generated population is provided. It is shown that such sketches are more similar to original photo images and their use in the discussed problem may lead to good results. But for all that, the created sketches meet the requirements of the truthful scenario as allow possibility of incomplete information in verbal descriptions. Results of experiments on CUHK Face Sketch and CUHK Face Sketch FERET databases and also open access sketches and corresponding photo images are discussed.