摘要: |
This technical note describes a method to compute a symmetry transform in 2-D images, based on matching image gradients around a center reference point. The method is particularly amenable to parallel processing, and we describe an implementation that runs on graphics processing units for significantly reduced execution time for VGA-sized images. The symmetry transform has potential for use in finding salient regions containing objects, and also may be applicable to stable object keypoints for recognition. |