The identifacation model of visual images using Berkeley Wavelet Pyramid based on fMRI

Abstract

Over the past decade, the research of decoding visual stimuli from brain signal recorded during a person viewing the image is developing fast as a challenging and up to date work in the neural decoding field. Previous studies have demonstrated that it's possible to identify the one of a set of natural images from human brain activity. In neural decoding studies, the computational model played a key role to establish systematic mapping between the visual stimuli and brain activity. Although Gabor Wavelet Pyramid (GWP) was widely used to construct the receptive field model, the model based on Gabor Wavelet Pyramid (GWP) is very complicated and has high computational complexity. In order to improve the efficiency of the process of image identification, here, a novel receptive field model based on the Berkeley Wavelet Pyramid (BWP) was used to build the relationship between fMRI activity and the visual stimuli. Compared with the receptive field model based on GWP, the BWT based model is localized in space and tuned in spatial frequency and orientation, and outperform the previous model in computational time. Moreover, our results showed that the BWT-based model is also able to identify visual images with comparable accuracy as GWT-based model.