Quantification and Relative Comparison of Synthesized Texture

Motivated by the desire to quantify the output of various texture synthesis algorithms, our method uses classical statistical measures to test the quality of outputs for five different categories of texture. We test our approach on three texture synthesis algorithms and present our results here. Of the three different algorithms tested, one is pixel based, the second a patch based approach, and the third a hybrid approach which combines the best features of the pixel and the patch based approaches. Our study clearly shows that none of the three approaches gives the ideal texture against which we compare the synthesized results. However, it does give a fairly decent measure to compare the three and choose one among them as the best.