Image Quality Assessment Method Based On Event-Related Potential

During the process of acquiring,compressing,processing,transmitting,and restoring visual information,various distortions are introduced.These distortions bring more or less obstacles in the processing,analysis and application of visual information.Therefore,in the view of accurate measurement of the perceived quality of visual information,a reliable and reasonable evaluation method needs to be designed,to obtain the best visual quality experience at minimum cost.The subjective quality evaluation observer in the traditional method is person,who gives subjective qualitative evaluation of the image quality.However,since human tends to give qualitative analysis with their perception of image quality,it is difficult to get precise quantitative scores.At the same time,because of great individual differences,the quality score obtained by traditional subjective methods can not reflect the actual quality perceived by human well.Therefore,this project intends to analyze the relationship between brain electrical changes and visual perception by directly measuring the brain electrical signals while people are watching different quality images.The brain electrical signals mainly reflect the different perceptual responses caused by image quality in human visual system,sensing changes caused by different degrees of distortion and differences in perception caused by different aesthetic levels.Based on subjective quality assessment criteria and related psychological experiments,we design an experimental paradigm that can reflect changes in perceptual quality and study the effects of image quality changes and different beauty images on EEG.The event-related potentials of EEG signals and the corresponding spatio-temporal characteristics are extracted to build a learning model between features and perceptual quality for obtaining more accurate perceptual quality scores.The main research contents of this article are as follows:(1)A distorted image quality assessment method based on induced P300 component was proposed.Different quality images are classified through the acquisition of EEG signals.Firstly,based on the Oddball experimental paradigm,the original and distorted images were used as stimulus respectively.Then,through the presentation of the random image sequence,subjects are given the distortion judgment of the image and the brain electrical signals are recorded at the same time.Finally,design the classifier for EEG signals to classify the EEG signals generated by different qualities to obtain evaluation scores for the quality of the images.The experimental results show that humans have different sensory responses to images of different qualities,and it is more direct and effective to use the method of measuring brain signals to evaluate image quality.(2)A method of aesthetic image quality assessment based on event-related potential difference was proposed.The difference of event-related potential caused by different aesthetic images was analyzed and the difference of aesthetic cognition process was explored.Firstly,based on the Go/No-go experimental paradigm,five groups of comparative experiments were designed to compare the event-related potential difference waveforms obtained from different beauty images.Then,the event-related potential difference was analyzed to obtain the image aesthetic evaluation.Finally,the EEG signals of different aesthetic images were classified and compare the results.Experimental results show that humans have different intensity perception responses to images of different beauty levels.Therefore,the perception of the image can be evaluated by the perception response.(3)A method for classification of EEG signals based on convolutional neural networks is proposed to construct a better mapping relationship between perceived EEG signals and the quality of images.Firstly,the deep network of multichannel sequence fusion is designed according to the characteristics of low spatial resolution and high temporal resolution of EEG signal.Then,the spatial-temporal characteristics of EEG signals are extracted using the hierarchical perception and interlayer fusion model of network.Finally,the mapping relationship between EEG signal features and perceived quality is established to realize the classification of perceived signals obtained from different quality images.The experimental results show that the proposed method can improve the classification accuracy of EEG signals to than the existing methods.