Research On JND Model Guided By Visual Saliency In DWT Domain

It has been known that human visual system(HVS)can be applied to describe the underlying masking properties for the image processing.In general,HVS can only perceive small changes in a scene when they are greater than the just noticeable distortion(JND)threshold.Recently,the cognitive resources of human visual attention mechanisms are limited,which cannot concentrate on all stimulus.To be specific,only more important stimulus will be reacted by the mechanisms.It can increase or decrease visual sensitivity,that is,visual attention has a modulation effect on the JND threshold.When it comes to visual attention mechanisms,we need to introduce the visual saliency to model the human perception more accurately.Visual saliency is the perceptual quality that makes an object,region,or pixel stand out relative to its neighbors and thus capture our attention.With a saliency map providing the information of locations where are visually salient to human visual system,region-based image processing can be performed more efficiently.Knowing the locations of the important regions broadly benefits applications such as region-of-interest image segmentation,object recognition,content-aware image editing and image retrieval,etc.The existing JND model rarely takes visual saliency into account,which leads to the JND model not being able to better integrate with the human visual system,which is not conducive to the development of some image processing.For this reason,we have innovated the JND model,combined it with visual saliency,and proposed a more optimized Saliency-Modulated JND(SJND)model.Discrete wavelet transform(DWT)is essentially a method of time-frequency analysis,which can perform multi-scale transform.The transformed image can obtain different resolution scales and image information at different frequencies,and has high-resolution characteristics.There are also excellent characteristics of crossover,which is more in line with human visual characteristics,so the models we propose are based on DWT.The main work of the paper is as follows:1.The JND threshold corresponds to the minimum value that the human visual system can perceive,and it can also be defined as the reciprocal of the contrast sensitivity function(CSF).The existing JND model generally consists of three parts: contrast sensitivity function(CSF),luminance adaptation(LA),and contrast masking(CM).Based on the framework of the original JND model,we mainly modified and innovated CSF,LA and CM,and optimized the corresponding parameters.Finally we proposed a SJND model with better effect.2,The application of visual saliency simplifies the cumbersome steps in image processing and greatly improves the work efficiency,and has been widely concerned.Therefore,it is especially important to combine it with the JND model in image processing.In this paper,an efficient saliency map detection method is used to modulate the JND threshold to form a new SJND model,which makes the thresholds of salient and non-significant regions different.In the image processing,different operations are performed on the salient region and the non-significant regionrespectively,which reduces image distortion and improves the perceived quality of the image.It should be noted that the modulation function of this paper is nonlinear,which can make the saliency map detection more consistent with the JND threshold.Finally,from subjective and objective experiments,whether it is the significant detection algorithm or the SJND model,our obtained results of the method is more superior.