Research On Image And Video Clearness Method In Inhomogeneous Medium

Image and video clearness technology has always been a research focus in computer vision and image processing.Due to the interference of the medium in the transmission process,the image and video have a certain degree of degradation,such as low contrast,dark colors and blurred texture details,which seriously affect the effect of subsequent processing,such as image interpretation,scene analysis,target recognition and other algorithms.Therefore,the image and video clearness technology in inhomogeneous media transmission has extensive research significance and application value.In this thesis,the main work and research content are as follows:(1)Research on the foggy image clearness based on convolutional neural network.Aiming at the problem of the current foggy image clearness algorithm can not be based on the scene to better distinguish the concentration of fog,resulting in partial image distortion.An outdoor foggy image clearness method based on convolutional neural network is proposed.The neural network is used to train the samples and learn the foggy image features,and the mapping relation between the foggy image and the transmission is established.At the same time,for the situation that the dark channel prior method is inaccurate for the estimation of the atmospheric light intensity of the image containing the sky region,this thesis proposes a self-adaptive estimation strategy of atmospheric light intensity based on the sky constraint.According to whether the image has a large range of sky regions,atmospheric light intensity is estimated in different ways.(2)Research on the single underwater image clearness based on color space.Aiming at the image distortion and noise amplification for general underwater image enhancement methods,and the traditional restoration method tends to make the image darker,this thesis improves the color space of non-local prior method.The phenomenon that the transmission estimation is too large in this method is corrected,and it can obtain better noise reduction effect and edge retention by increasing the limit of the edge points while solving the spatial constraint of transmission.And the background light is corrected by combining the dark channel prior and the color saturation,to obtain a more accurate background light.Experiments show that this method has better effect on underwater image clearness.(3)Research on the underwater video clearness based on spatial-temporal information fusion.In view of the phenomenon of flicker jumping in adjacent frames during video playback and the fact that some algorithms can not handle fast scene changes of the video well,a method of underwater video clearness based on spatialtemporal information fusion is proposed.Combining space and time information,estimate more stable transmission through interpolation and averaging,and reduce the flicker phenomenon due to the different background light value estimation of single frame image.The experimental results show that the method implemented in this thesis is good in the outdoor foggy image and underwater video clearness.At the same time,the continuity and smoothness of the video are also well reflected.