The Study Of Analysis Methods Of Images Emotion Based On FMRI

It is convenient for us to study brains by functional Magnetic Resonance Imaging(fMRI) which is invasive and high resolution. With the development of fMRI, it is used in the areas of medical and scientific. The study about fMRI data of the brain mainly focuses on the cognitive aspects, including identifying, attention and memory. Relatively,the study for the emotions is lesser.Especially the brain changes which images emotion aroused. With the development of technology,the appearance of computer and digital camera leads to the appearance of a lot of images.If we can store and retrieve these pictures better,this will provide more convenient to mankind. This paper changes from traditional methods which study images emtion from color and texture to the method which read the images emotions by fMRI.In order to read the images emotion better by fMRI,this paper does a deep study.Event-related fMRI design is used,which is appropriate for this paper.According to the experimental goal,this paper selects the stimulate materials which include positive,normal and negative by designed experiments. And this paper selects the appropriate subjects and sets the corresponding parameters.In the aspect of data analysis, there are four parts in this paper.It includes data preprocessing,feature selection, feature extraction and classification algorithm.The spm software is used to conduct data preprocessing;In the aspect of feature selection,this paper uses the methods of active region and t test.The experimental result shows that the method of t test is better;In the aspect of feature extraction,this paper uses five methods,including peak value, the time at the peak time, the change rate, the cumulated value and Principal Component Analysis(PCA).The experimental result shows that the time at the peak value is a more stable and accurate method in the five methods, and the other four methods are nearly; In the aspect of classification algorithms,this paper uses three algorithms, including svm algorithm, tuning svm algorithm and decision tree algorithm.The experimental result shows that tuning svm algorithm is more conducive to read out the three different emotions.The classification accuracy of this experiment is 78%, far more than the random level of 50%.This demonstrates that the selection methods in the experiment are feasible.These methods can predict the three emtions effectively.Therefore, it is feasible by fMRI to read out the images emotions in the brain. This paper has a certain reference and application value for the subsequent study of more data analysis methods and images emotions.