Research On Field-source-function In Stable Field Image Reconstruction

Digital images are one of the most important carriers for information recording and dissemination,but some information may be lost during the process of acquisition,editing,and transmission.Therefore,image reconstruction technology is needed to recover this lost information.Currently,the purpose of most image reconstruction research is to make it impossible for the human eye to detect,focusing only on the overall visual effect of the reconstruction results,rather than pursuing the accuracy of the reconstruction.However,with the widespread application of image analysis and recognition,unpredictable errors may occur when these reconstructed images are used for detail feature extraction.In addition,with the vigorous development of digital tools with image processing capabilities,reconstruction technology needs to meet the demand for real-time image processing.Therefore,researchers used stable field to describe local texture of an image,and proposed an image reconstruction model based on stable field to accurately and efficiently reconstruct missing information in damaged images.Based on the research of local texture stable field reconstruction model,this paper proposes effective field source selection and field-source-function construction algorithms based on the relationship between field and source in the model,aiming at the accuracy and efficiency issues of current image reconstruction algorithms.Firstly,the stable field model is combined with image texture,and the direction interval of the main texture direction at the defective pixel point is obtained based on the gradient information of each known point in the neighborhood of the defective pixel point.The known points within the interval are selected as effective field sources;Secondly,the differential approximation method is used for the effective field source,and the low order Taylor expansion is used to calculate the estimated value of the point at the defect point,thereby designing the similarity factor between the effective field source and the field generated at the defect pixel point,and combining the distance factor between the two to construct the field source function,finally completing the reconstruction.Comparative experimental results between the proposed algorithm and two improved algorithms based on typical reconstruction models,as well as two reconstruction algorithms based on stable field models,show that the peak signal to noise ratio(PSNR)improves by an average of 1.5 d B to 2.6 d B,the reconstruction accuracy improves by an average of 3.5% to 7.9%and the reconstruction efficiency improves by an average of 30 to 2000 seconds.Experimental results on selected images and Celeb A-HQ face dataset show that the algorithm achieves efficient and accurate reconstruction of different types of defective images.Based on the above research,further research was conducted on the local texture stable field reconstruction model of images,and it was found that current algorithms still exhibit defects such as structural discontinuity and erroneous extension when reconstructing defective areas with strong or complex structural features.To solve this problem,a local texture stable field reconstruction model of images based on structural tensor is proposed.Firstly,by introducing structural tensor eigenvalues to design structural data items and combining the confidence factor of defective pixel points,a priority function is constructed to specify the reconstruction order of defective pixel points in local defective areas.In addition,the image is divided into edge regions,texture regions,and flat regions based on the structural tensor coherence factor,and then the size of the fixed field source template is adaptively selected for the defective pixels located in different structural regions for subsequent reconstruction.Experimental results on the Places2 scene dataset and Celeb A-HQ face dataset show that the proposed algorithm improves the peak signal-to-noise ratio(PSNR)by an average of 1.5 d B and 1.8 d B,and the reconstruction accuracy by an average of 2.2% compared to the original algorithm.The proposed algorithm achieves better reconstruction visual effects and higher reconstruction accuracy for defective images with complex structural features and dense texture details while maintaining good reconstruction efficiency.