Nonlinear diffusion equation with a dynamic threshold-based source for text binarization

Binarization for degraded text images has always been a very challenging issue due to the variety and complexity of degradations. In this paper, we first construct a thresholding function for the input image in a local manner and then present an anisotropic diffusion equation with a source involving dynamic thresholding function. This dynamic thresholding function is governed by an auxiliary evolution equation, taking the constructed thresholding function as the initial condition. In the diffusion equation, the diffusion term achieves the edge preserving smoothing, while the source term is response for designating dynamically the text and background pixels as two dominant modes separated by the final dynamic thresholding function. To evaluate the proposed model solely, we only utilize the simplest finite differencing rather than more elaborated scheme to solve it numerically. Experiments show that the proposed model has generally achieved the superior binarization results to other nine compared models.