Application Reseach On Uncertain Information For Image Noise Processing

It is inevitable that various types of noises are introduced in the process of acquisition,transmission and recording no matter to macro image or to micro image.In this thesis,Noise removing in macroscopic image and micrograph and the location of fixed-point noises in microscopic image are all referred to as noise processing.Indeterminac prevails in the real world,and there are subjective uncertainties in human cognition.Just because these uncertainty factors of image,noise and human cognition,there are some shortcomings in existing noise processing algorithms.The main problems of traditional filters used in macro images and micro images are as follows:Lack of quantification and utilization of uncertainty information,poorer robustness,poor specificity of noise characteristics under high noise environment;poorer robustness of similarity measurement,sensitive threshold,and lack of objective evaluation of filtering performance used for microscopic video,etc.The main problem existing in the processing of fixed-point noise in micrographic is:Targets and fixed-point noises can’t be effectively separated for having no effective algorithm used for noise location.Therefore,this thesis quantifies the uncertainty information of image and noise from different aspects and uses the fusion information of uncertainty to solve the related problems above.Firstly,measures are reached by using image uncertainty information to improve the performances of the existing filtering used for the removal of random impulse,salt and pepper noise and Gaussian noise.Then,in neutrosophic image field,a novel algorithm based on double features is presented to locate the fixed-point noise in micrograph.Finally the proposed algorithms are used in the computer assisted sperm quality analysis system,and two novel quantification indexs are put forward.One is the quantification index for measuring of cleanliness of microenvironment and the other is a new objective evaluation index for measuring the performance of denoising algorithms used in microscopic video.Research contents of this thesis are as following:1)A new algorithm for the removal of random pulse are considered based on the fusion information of directional characteristics and uncertainty:For the sensitivity and uncertainty of noise threshold and the lack of direction information of ROLD(Rank-Ordered Logarithmic Difference)statistic,a new four-direction filter template is studied and a new noise statistic RODLD(Rank Ordered Directional Logarithmic Difference)is proposed firstly.Then a quantitative function of the uncertainty is constructed and the ROIN(Rank Ordered Neutrosophic Indeterminacy)statistic is proposed.In order to reduce the influence of sensitivity and uncertainty of noise threshold,a double-side noise detection scheme has been proposed by using RODLD and ROIN.To further improve the accuracy of weight function,a new bilateral filter is constructed by using ROAD(Rank-Ordered Absolute Differences)and IN(Neutrosophic Indeterminacy),and the similarity of pixels is measured in terms of gray and uncertainty.The experimental results show that the method based on orientation and uncertainty characteristics reduces the sensitivity of noise threshold.The performances of the proposed algorithm are much better than algorithms for comparison.2)A filter for removing salt and pepper noise based on the combination of Neutrosophic theory and Grey theory has been presented.To remove noise effectively and protect image details efficiently under high density environment,ECGCD(Extreme-Compression-Grey-Correlation-Degree)and RDD(Robust Dissimilarity Degree)features are presented to enhangce the distinctions of grey relational degrees in traditional grey filtering.RDD statistic has obvious characteristics of noise polarization and robustness,and can be used to detect noises effectively.A secondary detection algorithm based on fusion information of RDD and ROIN is constructed.Using ECGCD and NI to measure the similarity of pixels,an adaptive weight function based on the fusion of uncertain features has been established.Experiments show that the algorithm based on undetermined fusion information reduces the misjudgment rate of edge pixels,and improves the denoising performance greatly.3)A two-stage non-local means based on entropy neutrosophic indeterminacy is studied.In this paper the disadvantages of similarity measurement and filtering parameters in non-local filtering are discussed,and through the study of the properties of local entropy of image,a characteristic of describing uncertainty using truncated local entropy is proposed.It has strong abilities of reflecting the property of different regions and resisting noise.Therefore,the smoothing parameter is adjusted adaptively by using the entropy neutrosophic indeterminacy,and the weight function is constructed by combining the entropy neutrosophic indeterminacy and the gray characteristic to improve the accuracy of similar measures.Making full use of the useful information in the method noise,the new weight function and the pixel self-adaptive smoothing function are integrated into the NLM algorithm and a two-phase non-local filtering algorithm is proposed.Experiments show that the new algorithm has a good performance in the protection of image structure.4)An algorithm used for the localization of fixed-point noise in sperm microscopic video has been proposed:To overcome the defect of segmenting failure in traditional segmentation algorithms used for the extraction of sperm,a neutrosophic gradient feature that has the ability of noise adjusting has been designed.Therefore,an algorithm for the location of fixed-point noise based on two standard decisions has been proposed by combining neutrosophic gradient and neutrosophic gray characteristics.Experiments show that the proposed algorithm can successfully extract targets from an image without bimodal distribution and successfully separate sperms from noises.Evaluation of microenvironment cleanliness can be quantified by the new index RNSO(The Rate of Not Sperm Objects),and it fills the blank of the quality parameters used in microenvironment.5)In micrograph filtering,it is unsuitable using PSNR and SSIM indexs to evaluate the performance of a filter.STDS(Sum of Threshold Differences of a Sequence)has been creatively proposed as the evaluation index used to evaluate the performance of a filter in micro video.Compared with the comparative algorithms,the new algorithms have excellent STDS indexs.