An image binarization system for composite pictures
Y. Chigusa, T. Hattori, M. Ikegami, Mamoru Tanaka · 2003
The authors describe a novel binarization method based on neural dynamics for mixed gray-level and binary pictures. The Hopfield neural network is applied to this system. The threshold of each neuron is adaptively decided to be proportional to the gray-level of the corresponding input pixel, then the steady state of the network is assumed as the output image. The proposed algorithm is massively parallel. The two main advantages of this dynamic system are clearly established. One is that this system transforms composite pictures to pseudo gray-level pictures, without any segmentation. The other is that blurred images, especially in binary pictures, are reconstructed as sharpened pseudo gray-level pictures. The binarization method proposed is evaluated by using data obtained from real images and from the binary images reconstructed by the conventional binarization method.>