A Simple and Fast Method for Face Sketch Synthesis

Abstract

In this paper, we propose a simple and fast model-based face sketch synthesis method using subspace ridge regression. Different from the commonly used data-driven strategy that similar sketch patches are searched from the real sketch database, the proposed simple and fast face sketch synthesis (SF-FSS) method adopts the divide-and-conquer strategy which firstly classify the training photo-sketch patch pairs into a number of classes and then use a simple ridge regression model to learn the mapping between training photo patches and their corresponding sketch patches. We can predict sketch patches directly and fast by multiplying the input test photo patch with some special pre-computed regression matrix. Simultaneously, some high frequency information is also generated to be superposed on the regressed sketch patches. Experimental results show that the proposed method achieves comparable or better performance with least time consuming compared with state-of-the-art methods.