An Improved Image Thresholding Method Based On Two-Dimensional Histogram

The traditional two-dimensional histogram thresholding method ignores the existence of noise, which may lead to misclassification problem. Moreover, the calculation of the optimal threshold by two-dimensional Renyi entropy is too complex. To avoid these drawbacks, an improved segmentation method which combining grayscale-gradient histogram and traditional two-dimensional histogram is proposed in this paper. First, the gray-gradient histogram is constructed by using the gray value of each pixel and the gray value of its neighborhood. The noise points can be removed according to the distance between the pixels and the horizontal axis in grayscale-gradient histogram. After de-noising, the pixels are located near the main diagonal in the traditional two-dimensional histogram. Then, the distance from each pixel to the origin can be used to select the optimal segmentation threshold, which reduces the computational complexity from $O(L^{4})$ to $O(L^{2})$. The experimental result shows that the improved method proposed in this paper is faster, more accurate and more robust than traditional two-dimensional method, thus especially suitable for low signal-to-noise ratio (SNR) images.