Generation of Future image frames using Adaptive Network Based Fuzzy Inference System on spatiotemporal framework

摘要

This paper presents an algorithm for Future image frames generation using Adaptive Network Based Fuzzy Inference System (ANFIS) on spatiotemporal framework. The input to the network is a hyper-dimensional color and spatiotemporal feature of a pixel in an image sequence. The ANFIS is trained for R, G and B values separately for each and every pixel in image frame. Principal Component Analysis, Interaction Information and Bhattacharyya Distance measure have been used to reduce the dimensionality of the feature set. The resulting scheme has successfully been applied on satellite image sequence of a tropical cyclone. Two image quality assessment techniques, Canny edge detection based Image Comparison Metric (CIM) and Mean Structural Similarity Index Measure (MSSIM) have been used to evaluate future image frames quality. The proposed approach is found to have generated nine future image frames successfully.