VLSI curve detector

摘要

Line and curve detection find very important applications in image processing and pattern recognition. It is a key operation for line-like images and patterns. This paper presents a new VLSI architecture for implementing curve detection which has simpler control and interconnection structures and provides the capability to handle the backtracking in linear time. Firstly, we will propose a new curve detection algorithm which can be implemented by using the proposed VLSI architecture with extensive pipelining and parallel computation capabilities. The VLSI architecture proposed will find a curve of length n according to a certain figure of merit function in (n + min(M, N)) time units, where M × N is the number of pixels in the input picture and n is the length of the curve to be detected. It can then perform the backtracking procedure and determine the coordinates of the nodes along the detected curve in N + [log2M] + n time units. It will need O(N × M × n) time units, if using a uniprocessor. The algorithm partition problem is discussed. Verification of the proposed architecture is given. The important aspect of the curve detection algorithm-backtracking procedure is also studied in detail.