An adaptive split-and-merge method for binary image contour data compression

The split-and-merge method is a well-known algorithm for polygonal approximation in computer vision applications such as feature extracting and pattern matching. Its accuracy depends on the tolerance, that is the error threshold value. This study presents a split-and-merge method with an adaptive tolerance value for compressing image contours. The tolerance value, which depends on the grid constant D and the line length of line L in a collinearity test, is adopted to reduce quantization error while keeping its original shape. A contour tracing method that achieves the right shape representation of binary images is also discussed. Experimental results for real binary contours show the method is effective for compression of a binary image. The proposed method allows a precise description of the original image and can smooth coarse contours. It is also computationally efficient.