Complexity Analysis Based on Image-Processing Method and Pixelized Recognition of Chinese Characters Using Simulated Prosthetic Vision

Abstract

The influence of complexity and minimum resolution necessary for recognition of pixelized Chinese characters (CCs) was investigated by using simulated prosthetic vision. An image-processing method was used to evaluate the complexity of CCs, which is defined as the frequency of black pixels and analyzed by black pixel statistic complexity algorithm. A total of 631 most commonly used CCs that can deliver 80% of the information in Chinese daily reading were chosen as the testing database in order to avoid the negative effect due to illegibility and incognizance. CCs in Hei font style were captured as images and pixelized as 6 x 6, 8 x 8, 10 x 10, and 12 x 12 pixel arrays with square dots. Recognition accuracy of CCs with different complexity and different numbers of pixel arrays was tested by using simulated prosthetic vision. The results indicate that both pixel array number and complexity have significant impact on pixelized reading of CCs. Recognition accuracy of pixelized CCs drops with the increase of complexity and the decrease of pixel number. More than 80% of CCs with any complexity can be recognized correctly; 10 x 10 pixel array can sufficiently provide pixelized reading of CCs for visual prosthesis. Pixelized reading of CCs with low resolution is possible only for characters with low complexity (complexity less than 0.16 for a 6 x 6 pixel array and less than 0.24 for an 8 x 8 pixel array).