Research On Image Enhancement Technology Based On Deep Learning

Nowadays,more and more computer vision systems are used in various industries.Most of these computer vision systems work on the premise of inputting clear images.However,actual scenes,especially in outdoor environments,due to weather conditions such as fog and haze,we cannot guarantee to obtain clear and usable image data.Therefore,image enhancement technology has attracted more and more attention from researchers.Meanwhile,with the improvement of science and technology,the development of the ocean has attracted more and more attention from all countries.The application of underwater robots for underwater detection is considered to be an important task,and the application of computer vision is a key part of accomplishing this task.However,light absorption and scattering phenomena also exist in the underwater environment.Compared with foggy images,underwater images will suffer from more serious distortion problems,such as reduced contrast and too much blue-green.In view of the impact of outdoor environments such as haze on image clarity,dehazing image enhancement technology is considered to be an effective method to solve this problem.Haze removal typically works on a physical model to estimate how light is transmitted and lost due to absorption and scattering through the atmosphere.In this paper,a region detection network is proposed 'to learn the relationship between hazy image and medium transmission map in a patch-wise manner;the transmission map is then used to remove haze via an atmospheric scattering model,and enhance the detail of de-hazed images.The model is mainly composed of two basic network units and can be trained in an end-to-end manner.One network unit is a module with the residual structure that facilitates the learning process of deep network.The other is a novel module with cascaded cross channel pool,which fuses multi-level haze-relevant features and boosts the abstraction ability of the model on a nonlinear manifold.Moreover,an evolutionary-based enhancement method is developed to improve the level of detail of over-smoothed results.Several comparative experiments have been conducted on synthetic and real images,through which we conclude that the proposed method achieves state-of-the-art haze removal results,qualitatively and quantitatively.Moreover,we present a lightweight version of the proposed network,which achieves an impressive haze removal performance even on low-power devices.To improve the visual quality of underwater images,we proposed a novel enhancement model,which is a trainable end-to-end neural model.Two parts constitute the overall model.The first one is a non-parameter layer for the preliminary color balance,then the second part is consisted of parametric layers for a self-adaptive refinement,namely the channel-wise linear shift.The neural network used is a joint optimization model,which includes supervised learning and unsupervised learning,where supervised learning is used to align the label value and the predicted value at the pixel-wise level,and unsupervised learning is to further ensure the enhanced image quality.The experimental results show that the model can solve the problem of underwater image color cast and color unsaturation caused by limited lighting conditions.