Partialdifferential equation based
ImageEdge Detector and Enhancement
Earlier Work
We were among the very firstwho proposed the use of fourth order and other higher order differentialequations for image processing, see our paper. The fourth order operator, when combinedwith appropriate nonlinear production (or the so called enhancement) terms,delivers better results for image restoration (including deblurring andenhancement).
Recent Work
Albeit anisotropic diffusiontype of PDEs are very good at image denoising and deblurring, they are not asgood as other standard techniques for image edge detection. In particular, edgedetction of texture images is still a challenging task for all exist methods. Clearly, a different approach is required. We borrow the idea ofsynchronization from nonlinear dynamics. Two coupled nonlinear PDEs are set toevolve with different time scales. Image edges are obtainedthe synchronization residual. Our results are compared with those obtained by using the standardCanny and Sobel detector. For each original image, one optimal result ispresented in this page. A large grayscale image can be viewed by clickingon the image. For details, see our paper.
Edge Detection Edge Detected Images by couple PDEs |
Edges Detected by Sobel Edge Detector | |
Edges Detected by Canny Edge Detector | |
Image Enhancement Great edge imag edetection leads to great enhancement, which is obtained by adding the edge to the original image. Edge Enhanceed Images by coupled PDEs |