Infrared Image Understanding Based On Bionic Vision Computation Models

With the rapid development of computer and infrared technology,infrared image understanding has been a specialized research direction and application way.It integrates computer science,mathematics,bionics and infrared imaging,is the important content of the current infrared visual research.For the infrared mage scene,the study based on the bionic visual computing model is the current trend of infrared image understanding.By studying different visual computing models,Infrared image understanding,not only can reduce the redundancy of information processing,but also can improve the perception ability.This efficient way of information processing has very important theoretical significance and applicable value in the civilian and military field.This paper mainly carries out the next research work,following the infrared image understanding based on the bionic vision computing models.(1)Proposing infrared image enhancement vision computing models based on hierarchy and Cellular Automata to enhance infrared images.Traditional image enhancement computing models cannot improve the contrast of the infrared image effectively,lose necessary image structure information,and make the boundaries of the object smooth.This paper presents two infrared image enhancement vision computing models,named infraraed image enhancement vision computing model based on hierarchy and infrared image enhancement vision computing model based on Cellular Automata,respectively.These two computing models apply hierarchy and Cellular Automata explore the structure data,achieving effecive infrared image enhancement results.(2)Proposing saliency detection vision computing models based on Boolean map to estimate salient regions.Traditional Boolean-based saliency detection models can estimate saliency in simple scenes,but less accuracy in the complex infrared image scenes.This paper presents two saliency detection vision computing models based on Boolean-based saliency by Bayes and graph theory,named saliency detection based on Boolean map and foreground map and saliency detection based on graph-Boolean.These two vision computing models analyze the characteristic of Boolean map,and apply vision information,achieving the saliency detection in the complex infrared image scenes and natural scenes.They can suppress the background noise effectively,and improve the accuracy of detection.(3)Proposing infrared object recognition vision computing model based on Locally Adaptive Regression Kernel(LARK)to locate Infrared objects.Traditional LARK-based object recognition models show good performance in the natural scenes,but less accuracy in the complex infrared scenes,losing structure information easily.This paper presents two object recognition vision computing models based on LARK features,named infrared object recognition vision computing model based on local and global LARK features and infrared object recognition vision computing model based on LARK features and Booean map.These two vision computing models analyze the characteristic of LARK features,and apply heat equation and Boolean map respectively,leading an improved object recognition result,without any learning knowledge.(4)Proposing infrared image super-resolution vision computing model based on transformed self-similarity for infrared image super-resolution.Due to losing texture details and structure information in low resolution infrared image,super-resolution models often restore low quality results.This paper presents an infrared image super-resolution vision computing model based on transformed self-similarity,which analyzes the structure details by applying appearance features,region covariance,dense error and scale information,retains rich image details,leading an improved super-resolution accuracy.