Why Does the Sketch Look So Vivid?

As an effective and convenient intermediate means, sketch has been playing an important role in expressing human intentions. Aware of its significance, plentiful sketch related studies have been conducted, such as sketch based image retrieval and recognition. However, very few works have researched the intrinsic factors that make sketch look vivid. In this paper, we explore the key to a good sketch by a weak supervised approach to discovery discriminative patches for different categories of sketches. For each class, we randomly extract tens of thousands of patches at multiple scales, which are represented by Pyramid Histogram of Oriented Gradient (PHOG). Then all patches are thrown into an iterative detection process to find the most discriminative ones. Experimentally, we analyze the visual results and give a reasonable discussion about good and bad sketches. Furthermore, more experimental results on the TU-Berlin sketch benchmark dataset demonstrate the effectiveness of the proposed method, as compared to other available approaches.