Research On Inverse Halftoning Based On Machine Learning

Halftone image is the kind of digital images meeting to bi-level output and display equipment,which is widly used in the printing industry,desktop publishing systems and light emitting diode(LED)display,etc.When viewed from a distance the halftone image can resemble its original grayscale image.Since halftone images have only 1-bit binary such as 0 and 1,the common image processing approaches,such as rotation,resizing,feature extraction or segmentation,etc.,cannot directly be used in halftone images.Thus,how to realize inverse halftoning is a promising but challenging research area in halftone transform,digital archive management and high precision identification of halftone.For the inverse halftoning,the main contents are carried out as follows:1.In order to restore high quality continuous tone images from each class of halftone images,halftone image fine classification is the foundational work.Combined with deep learning a stacked sparse auto-encoders deep neural network halftone image classification algorithm is first proposed to address the fine classification for 14 kinds of halftone images in four types of ordered dither,error diffusion,dot-diffusion and direct binary search.In order to reduce the run-time of deep neural network and improve the image correct classification rate,an effective patch extraction method is proposed for testing halftone images by using the mean and variance of local entropy in a patch.The experimental results demonstrate that both the correct classification rate and the types of classified images of the proposed method are higher than the state-of-the-art methods' on two public testing image sets.2.To exploit the correlation between the color components of a halftone image in digital color inverse halftoning LUT method the extreme learning machine(ELM)is adopted to estimate the nonexistent values in color table according to all existent values in the same patterns of three color component tables.ELM applied to construct the color LUT tables not only can improve the fitting precision of nonexistent values but also does not reduce the image transformation speed of LUT inverse halftoning algorithm as its extremely high operational efficiency.Comparing the template selection color inverse halftoning method the proposed algorithm can eliminate most of the noise generated by errors of fitting nonexistent values in converted continuous-tone images.3.Aiming at the nonlinear mapping model construction for heterogeneous image modalities a novel semi-coupled multi-dictionary learning inverse halftoning method based on structural clustering and sparse representation is proposed,which could address the crossstyle image restoration from halftone images to continuous-tone images.The learned semi-coupled multi-dictionary pairs can well represent the structure characteristics of halftone images and continuous-tone images,respectively.In addition,the mapping functions learned by semi-coupled manner can bridge the gap between the two different style images of halftone image and continuous-tone image.Unlike the coupled dictionary learning methods,the proposed method could effectively relax the assumption of the same sparse coding coefficients in coupled dictionary learning and obtain more accurate mapping functions.The experimental results demonstrate that the proposed method can restore higher quality continuous-tone images than that produced by the state-of-the-art methods,which not only reduce the screen noise in smooth regions,but also provide well fine details and clear edges.4.For the problem of the halftone image restoration model construction and optimization,a evolutionary deep learning algorithm for inverse halftoning is proposed which directly learns an end to end mapping from the halftone images to continuous-tone images.This is first introduced the deep convolutional network into the inverse halftoning.Through training the designed deep CNN model all the model parameters of the feature extraction of halftone,the nonlinear feature mapping from halftone to continuous-tone and image reconstruction are jointly optimized.On this basis the residual learning strategy is introduced to train another deep CNN model to remove the remaining dot and artificial feature from the preliminary restored continuous-tone image in order to reconstruct a high quality continuous-tone image.Our extensive experiments demonstrate that the performance of the proposed method is improved significantly comparing with the published methods' on both quantitative and qualitative evaluation.