Research On Inverse Halftoning Method Based On Extreme Learning Machine And Look-up Table

In the process of printing,the digital continuous tone image will be halftone processing before printing.The converted halftone image has the same visual effect as the origin al image,and solves the problem that the traditional printing can not achieve continuous tone image printing,but the halftone noise is introduced,which causes the missing information of the original image to a certain extent.If you get a halftone image and you want to perform digital image processing operations such as compression,scaling,enhancement,and recognition,you must first perform inverse halftoning processing on the image to restore the missing part of the original image.At present,there are many methods for inverse halftoning of digital images.The look-up table(LUT)inverse halftoning method has great advantages,such as simple calculation,low complexity,easy programming and parallel processing,etc.In this paper,we study the two main factors affecting the quality of LUT inverse halftoning image,null estimation and template selection,and achieve inverse halftoning of gray-scale and color halftone images with better objective effects and subjective effects.The main research work is as follows:(1)Aiming at the problem of fitting the null term in LUT,a LUT inverse halftoning null estimation algorithm based on extreme learning machine(ELM)is proposed.In the initial stage of establishing the LUT,because the continuous values corresponding to all the index values are not extracted in the training sample,the LUT is incomplete and the null value appears,and the accuracy of the null value directly affects the quality of the inverse halftoning phase inverse halftoning image.The algorithm of null value estimation based on ELM presented in this paper has the advantages of simple and easy operation,high calculation speed and high estimation accuracy.(2)A LUT template intelligent optimization method based on Particle Swarm Optimization(PSO)is proposed.The optimal template obtained by PSO not only has a relative position relation,but also adds a relative order,and at the same time better reflects the correlation between the image pixels and their neighboring pixels.(3)For the poor smoothing effect of LUT inverse halftoning image,an inverse halftoning method combining LUT and Gaussian filtering is proposed.The image fusion of the results of improved LUT and Gaussian-filtered inverse halftoning methods is applied to inverse halftoning processing of gray-scale halftone images and color halftone images.Experiments show that the image after image fusion is better than the image after traditional inverse halftoning processing in objective evaluation and visual effect,and is more similar to the original continuous tone image.