Hierarchical Oil Painting Stylization with Limited Reference via Sparse Representation

submitted to 2015 IEEE International Workshop on Multimedia Signal Processing (MMSP)

• Saboya Yang     yangsaboya@pku.edu.cn
• Jiaying Liu         liujiaying@pku.edu.cn
• Shuai Yang        williamyang@pku.edu.cn
• Sifeng Xia          xsfatpku@pku.edu.cn
• Zongming Guo  guozongming@pku.edu.cn

Experimental Results

In the experiments, we compare the proposed algorithm with Zhao's method [4] and the BrushStroke APP [5].

To ensure the credibility of our method, we invited 30 testees with different ages, different genders, and different backgrounds, to finish the survey we made. In the survey, we ask testees to score the similarity of the style between the source image and the stylized images created by the mentioned three methods from 1 to 5 respectively. To avoid the testees guessing which is our method, the orders of three methods are randomly changing every round. As shown in Table 1, while 5 refers to the most similar one, our method acquires the highest score in each round.

Table 1: Scores of Different Methods in Images

Images	Proposed	Zhao's method [4]	Brushstroke[5]
1	3.50	2.21	3.29
2	4.04	2.79	2.75
3	3.11	2.36	2.29
4	3.67	1.97	2.90
5	3.33	1.47	2.77
6	4.07	2.40	2.60
7	3.57	1.83	3.30
Average	3.61	2.15	2.84

We also ask the testees to choose which creates the most similar one among the three methods in each round. And Fig.2 shows that more than 60% testees think that ours is the best for each round. The average result is 76.79% which demonstrates that our proposed method outperforms the other two methods in most people's eyes.

Fig.2 Statistics of best method Selection with different images.

In fact, both Zhao's method and the BrushStroke APP can only convert the target image into the general oil painting style without considering the individual drawing style of the source stylish image. But the proposed method is applicable to the oil painting stylization problem with limited reference, which is more useful. Moreover, experimental results demonstrate that the proposed method produces better stylized images than these aforementioned methods do.