Fig nbsp xA Non local cost aggregation for pixel p

Fig. 7. Non-local cost aggregation for pixel p on the cross-trees structure with a prior. Green arrows represent cost aggregation flow on horizontal scanlines and red arrows represent cost aggregation flow on the vertical scanlines. A cross-based structure which consists of these color arrows is the support region of pixel p. (For interpretation of the references to color in this SBI-0206965 figure caption, the reader is referred to the web version of this article.)Figure optionsDownload full-size imageDownload high-quality image (117 K)Download as PowerPoint slide
To a certain extent, our work is also motivated by the cross-based structures methods [19] and [20]. Instead of estimating the cross-based structures locally, we use the priors to locate potential local support regions? boundaries and then estimate the support regions as cross-structures on the horizontal and the vertical scanlines implicitly by cutting the cost aggregation flow as shown in Fig. 7.
Note chorion for the superpixel prior, there are over-segment boundaries even between neighboring pixels within low-textured regions. Such over-segment boundaries will not terminate the cost aggregations from further pixels on the path since the edge weights (equal to pixel similarity as defined in (7)) are still low.