Research Of Color Image Enhancement Based On Feature Preservation

Image enhancement technology is an important branch in image processing,and its purpose is to selectively highlight or preserve important features of the source image by adding specific information or making specific transformation,so as to obtain the resulting image that meets the needs of the users.With the rapid development of computer vision technology and the continuous improvement of people's requirements for image quality,image enhancement technology involves more and more research fields.Because color image is very popular now,and its content and expression of rich information is incomparable to grayscale image,therefore,the study of color image enhancement technology is particularly important.At present,recolorization and decolorization are hot issues in the field of color image enhancement.In this paper,we first reviewed the status of the relevant research,and then we carried out deep research and discussion on some key issues of feature preservtion in recolorization and decolorization.The main research results are presented as follows:(1)Classical recolorization methods use a total variation(TV)regularizer to preserve the details and suppress the noise of the resulting images.These methods can sometimes cause staircase effect and geometrical structure details over-smoothed.To address these problems,we propose a new recolorization method(ATGVR method)in which an adaptive second-order total generalized variation(TGV)regularizer is designed.Here,the adaptive second-order TGV regularizer is a weighted second-order TGV regularizer.The weight is computed by an adaptive edge indicator function.In addition,an efficient numerical method is developed to solve our new model.The numerical method is based on the weighted primal-dual algorithm and the forward-backward splitting algorithm.Experimental results and comparisons demonstrate that our newly proposed ATGVR method can generate better results than classical TV regularizer-based methods in the aspects of the inhibition of staircase effect and the preservation of image details.And the resulting images obtained from our newly proposed ATGVR method are more realistic and clear than those obtained from the classical TV regularizer-based methods.(2)The color informations of the target image can easily interfere with the process of recolorization in traditional recolorization methods,so the color style of the resulting image will be affected.To address this problem,a new recolorization method(CFTVR method)based on central constraint is proposed.On the one hand,a new centralized constraint term is designed in our new recolorization model,so as to reduce the interference of the color informations of the target image to the process of recolorization,and make the resulting image better preserve the color informations of the reference image.On the other hand,a fractional-order total variational regularization is introduced into our new recolorization meodel to more easily suppress the staircase effect and protect the resulting image details.In addition,in order to solve the new model,an efficient numerical method is designed by combining the augmented Lagrange algorithm and the primal-dual algorithm.Experimental results and comparisons demonstrate that,compared with traditional recolorization methods,our newly proposed CFTVR method can not only make the color style of the resulting image more faithful to the reference image,but also achieve better recolorization effect.(3)Since there are the low-efficiency preservation of the layering and details of the original fabric image,and the high-complexity algorithm in traditional fabric image recolorization method,an improved fabric image recolorization method(SDFR method)based on significant color extraction and image decomposition is proposed.On the one hand,by employing the significant color extraction,more representative main colors and segmentation regions of the original fabric image are obtained to make the resulting image better preserve the layering of the original fabric image.On the other hand,image details of the resulting image are better preserved by using the image decomposition and anti-overflow recolorization strategy.Experimental results and comparisons demonstrate that,compared with traditional fabric image recolorization method,our newly proposed SDFR method can not only effectively preserve the layering and details of the original fabric image,but also significantly improve the operation efficiency.Our newly proposed SDFR method is simple and efficient,and can be better applied to the color design work of fabric images.(4)Since there is the low-efficiency preservation of the contrast,the artifact problem and the high-complexity algorithm in traditional decolorization methods based on bimodal gaussian,a new real-time decolorization method(GBGD method)is proposed by combining gradient preservation and global contrast preservation.Firstly,to effectively preserve the local contrast of the grayscale resulting image,a new gradient preserving constraint is designed by using the gradient of the grayscale resulting image and the original color image gradient controlled by the brightness.Secondly,the global pixel contrast is introduced into the constraint of bimodal gaussian function to better preserve the global contrast of the grayscale resulting image.In addition,in order to solve the new model,an efficient numerical method is designed by combining the fixed-point iteration algorithm.Experimental results and comparisons demonstrate that,compared with traditional decolorization methods,our newly proposed GBGD method not only can preserve the contrast of the original color images and suppress artifacts effectively,but also has better performance in the evaluation index.Our newly proposed GBGD method can achieve real-time speed and has good application value.(5)Since there are the low-efficiency preservation of the main visual contrast,the layering and the details of the original color image in traditional decolorization methods,a new discrete search decolorization method(LLEGSD method)is proposed by combining local linear embedding theory and gradient similarity measure.On the one hand,the local linear embedding idea is introduced to make the grayscale resulting image keep the main visual contrast and the layering of the original color image.On the other hand,the gradient similarity measure is used to make the grayscale resulting image keep the edge details of the original color image.Finally,the grayscale resulting image is obtained quickly by using the discrete search strategy.Experimental results and comparisons demonstrate that our newly proposed LLEGSD method can not only effectively preserve the main visual contrast,layering and edge details of the original color image,but also achieve better decolorization effect than traditional methods.