A fast general algorithm for extracting image features on SIMD mesh-connected computers

Extracting features of components in an image is an important step for recognition of objects in the image. It is an intermediate-level image processing task as it transforms an image to a list of features. SIMD mesh-connected computers are considered natural architectures for low-level image processing tasks that transform an image into another image. In this paper, we design a fast general algorithm for extracting geometric features of image components, demonstrating that SIMD mesh-connected computers can also perform well on intermediate-level image processing tasks. We observe that a geometric feature of an image component is the combined contribution of the individual pixels in the component. We rst develop a general formula for extracting some geometric features of image components. A speciic geometric property of components, such as area, perimeter, compactness, height, width, diameter, moments, and centroid, can be computed by the general formula with a weight image specifying the contribution of each pixel to the property of the component it belongs to and an associative and commutative operator specifying how to combine the contributions. We then design a fast algorithm for the general formula on SIMD mesh-connected computers. The algorithm uses a pipelining technique to compute a geometric property of kn components at the same time. It takes O(d m kn en) time to compute a geometric property of all the components in an n n image on an n n SIMD mesh-connected computer with O(k) local space in each processing element, where m is the number of components in the image and k is an integer between 1 and d m n e, depending on the local memory available in each processing element. Compared with previous algorithms, it is simple and speeds up the feature extraction process by at least k times. Abstract { Extracting features of components in an image is an important step for recognition of objects in the image. In this paper, we develop a general formula for extracting some geometric features of image components such as area, perimeter, compactness, height, width, diameter, moments, and centroid. We then design a fast algorithm for the general formula on SIMD mesh-connected computers. The algorithm uses a pipelining technique to compute a geometric property of kn components at the same time. It takes O(d m kn en) time to compute a geometric property of all the components in an n n image on an n n SIMD mesh-connected computer …