Research On Infrared Image Enhancement Algorithms

Infrared imaging technology has been widely applied in reconnaissance warning,public security and medical treatment since its can well work in full-time,concealed himself and against interference.However,it has inherent deficiency that leads to the image captured by infrared camera with lower contrast,less edge detail,lower signal-to-noise ratio and poor visual effect than visible image;which will affect the performance of feature extraction,object detection and tracking.In order to make full use of the advantages of infrared imaging technology,it needs to enhance the infrared image which will be well suitable for human visual system.Thus,infrared image enhancement is very significant for enhancing the quality of infrared image.The subject of this thesis comes from the collaborative research project,which is mainly researched on infrared image enhancement technology.Based on the characteristics of infrared image and classical enhancement algorithm,combined with the theory of stationary wavelet transform and image super-resolution enhancement,the research is realized.The corresponding improved algorithm improves the implementation effect and image enhancement quality of the infrared enhancement algorithm.This paper mainly consist of the contents as follows:Aiming at the problem of blurred edge contour,low contrast and poor detail ability of infrared image,this paper designed the image denoising algorithm based on non-subsampled contourlet transform.It leverages the multi-directional sub-band to recognize noise.Meanwhile,a perfect nonlinear enhancement function is also used to enhance the detail coefficients.For this algorithm,it is robust to noise but very time consuming.To address this,a new infrared image enhancement algorithm based on stationary wavelet transform is designed,which introduces a saliency map to improve the performance and efficiency.Experimental results shown that the designed algorithm has better visual effect and higher efficiency than the traditional infrared image enhancement based on non-downsampled contourlet transform.Aiming at infrared image super-resolution,this paper designed an infrared image enhancement algorithm based on convolutional neural network.It performs SRCNN based super-resolution for the enhanced sub-bands of stationary wavelet transform,which can enhance the detail information of an infrared image.For the real-time problem of infrared image enhancement,the residual structure and batch normalization are introduced.Based on these,an improved FSRCNN is designed to achieve infrared image enhancement in real time.As to the analysis above,we performed the designed algorithms by MATLAB.And some widely used subjective and objective evaluation metrics to evaluate the performance of the designed and compared algorithms.The experimental results show that the designed infrared image enhancement algorithms can significantly improve the image quality,and the efficiency,which basically meets the real-time requirement.