Image Retrieval Technology Based On Electroencephalogram

The technology of image retrieval is often used to identify and filter special images.Compared with human vision(HV),machine vision(MV)is susceptible to factors such as complex image background and special semantic gap,therefore,the MV is difficult to obtain accurate image recognition results,and the adaptability is poor.The HV is a result of human long-term evolution,and its information processing channel has robust and reliable characteristics.Therefore,the integration of human brain intelligence and machine intelligence for image retrieval is a new research hotspot in this field.At present,the EEG-based image retrieval technology mostly uses rapid serial visual presentation paradigm.The RSVP paradigm attempts to detect the specific P300 component induced by the target image for to achieve the rapid detection of target images.Thus,P300 component detection methods are the core and key of EEG-based image retrieval technology.This paper will focusing on the analysis of P300 components changes,and on this basis,study how to further improve the P300 component detection accuracy and efficiency from detection algorithm,experimental paradigm and other aspects,thereby enhancing the performance of EEGbased image retrieval.This paper's main work is as follows:1.Aiming at the problem of P300 component latency jitter,the detection algorithm based on spatio-temporal filter is a more effective method.The representative is the HDCA algorithm and the sHDCA algorithm.However,the HDCA algorithm is use a fixed-time window strategy,so it is difficult to adapt the latency changes of P300 component.And the sHDCA algorithm is use a sliding window strategy,although can be better adapted to P300 latency jitter,but its calculation is more complex,and cannot be applied in real-time systems.To overcome the problem,a folding hierarchical discriminant component analysis(fHDCA)method is proposed to self-adapt the temporal variability.The innovation of the design is introducing information of prior moments into the current time window through a method of folding.The results demonstrated that the proposed fHDCA and sHDCA both perform significantly better than the standard HDCA algorithm in RSVP based target image detection task.Moreover,under the same accuracy,the time complexity of the fHDCA algorithm is one magnitude lower than that of the sHDCA,where the computing time is same level HDCA algorithm.2.In the process of analyzing the cause of the P300 latency jitter with different complexity image,we find that there is a difference in the brain response pattern and the P300 component induced by the image.The P300 component induced by highly complex image has a longer latency relative to the simple image.Thus,we proposed a novel method,Target Recognition using Image Complexity Priori(TRICP)algorithm,in which the image information is introduced in the calculation of the interest score in the RSVP paradigm.This method evaluates the complexity of the stimulus image through machine vision,predicts the possible range of P300 component latency,and adjusting the classifier to achieve better detection accuracy.The experimental results show that the method can adapt to the P300 latency jitter due to the image complexity change,therefore,the accuracy of the target detection can be further improved.And the method provides a new way to integrate human intelligence and machine intelligence.3.Compared with the single-trial P300 component detection,the multi-trial P300 detection algorithm has better performance.However,in target image detection application,obtaining images repeatedly is inappropriate for the RSVP paradigm because this method is time consuming and unconducive to real-time target detection.To solve this problem,this paper verified the feasibility of the dual-RSVP paradigm proposed by Cecotti in EEG data and further proposed a triple-RSVP paradigm.Experimental results show that the dual-/triple-RSVP can effectively improve the accuracy of target recognition.On this basis,this paper build the multiple RSVP framework for target image retrieval,and analyzes its advantages and limitations,and provides a new experimental framework for EEG target image retrieval technology.