An Efficient CannyRaw Edge Detection Algorithm for Raw Images

With the considerable increase in popularity of mobile photography, a large portion of efforts has been made to optimize the image signal processing (ISP) pipeline that transfers the raw images obtained directly from mobile camera sensors into color images. In this work, we demonstrate that the raw images are quite informative by showing the edges extracted from raw images and color images across a bunch of edge detection algorithms. Given that the characteristics and distribution of raw and color images are different, we present CannyRaw, an efficient edge detection algorithm designed for raw images, requiring no prior knowledge of the sensor and optics. Experimental results of the edge images under different illumination levels illustrate the robustness of CannyRaw to illumination and noise. Besides, the system power and resources can be reduced by the elimination of ISP.