Research On Specific Image Preprocessing Algorithms For Uneven Brightness Distribution

For some uneven brightness distribution images,effective preprocessing is conducive to subsequent processing.Both low-illumination images and auroral images are images whose luminance distribution is concentrated in the lower interval.This thesis focuses on low-light image,auroral image and corresponding pre-processing methods,and improves existing algorithms according to actual conditions.First,the low-light image is studied,and a low-light image enhancement method based on the logarithmic image processing model is proposed.The algorithm first divides the low-illumination image into the low-light image and the locally highlighted low-illumination image according to the brightness distribution.For the low-light image,firstly,a logarithmic image processing model-based enhancement is adopted to enhance the overall brightness.Secondly,the local illuminance information is used to perform adaptive illumination compensation based on the logarithmic image processing model.For locally highlighted low-illumination image,the image is divided into dark region and highlighted region by using the improved gravitational search algorithm according to the V components in the HSV color space.Secondly,according to the principle of maximizing brightness standard deviation,enhancement for different regions is performed respectively.After that,the enhancement results are fused using contrast pyramid-based method to obtain a balanced enhancement effect.Finally,the image illumination estimation and adaptive illumination compensation are performed to enhance the local brightness of the image.The experimental results show that the proposed algorithm can effectively enhance the brightness of low-light images with low noise amplification and color distortion,and enhance the brightness of dark region and the contrast of highlighted region in locally highlighted low-illumination images.Traditional quantization methods can lead to information loss,noise amplification and unclear texture in the preprocessing of auroral images.In view of this situation,this thesis proposes a nonlinear quantization method for auroral images according to statistical properties.Firstly,the gray distribution of a large number of auroral images is analyzed and found similar to Inverse Gaussian distribution.Then,the morphological characteristics and energy distribution of auroral images are analyzed in detail,and the gray distribution intervals of noise,background,texture and highlight structure in auroral images are precisely divided.Finally,an s-shaped nonlinear quantization function is designed according to the statistical and structural features.The experimental results show that compared with the traditional methods,the auroral images preprocessed by the proposed method have clearer texture,purer background and less noise.And the classification accuracy reaches 90.85 %,which is 1.85%-8.65% higher than the traditional method.