Color Image Enhancement Algorithm Based On Edge Preserving Filter

In image processing,image enhancement is an indispensable technique as an early stage of image preprocessing.The purpose of image enhancement is to highlight the details of the region of interest in the image,filter out the noise,expand the difference between the different objects in the image,improve the visual effect of the image and make the enhanced image more suitable for human visual system and the identification system of the machine.In recent years,with the development of science and technology and the continuous improvement of human needs,color imaging equipment is developing rapidly and being widely used in many aspects of life and work.However,usually the visual effect of the image in the hardware acquisition and imaging process is easily affected by the digital image capture device,the outside conditions of the scene and other factors,resulting in the degraded image which is even hard to recognize for human and machine analysis.Therefore,the research of color image enhancement technology has attracted much attention.Based on the requirements of the practical application,with studying the relevant theories of color image enhancement and analyzing the existing problems in color image processing,the paper proposes several new enhancement algorithms based on the characteristics of edge preserving filter.The new enhancement algorithms mainly includes color image enhancement based on quaternion guided filtering,color image enhancement based on multi-scale guided filtering and color image enhancement based on depth bilateral filter.As the field of color image enhancement is more extensive,the three color image enhancement methods proposed in this paper are for different application requirements.To make the detail of color images outstanding,highly visible and avoid color distortion in the process of the image enhancement,an image enhancement method based on the quaternion guided filter was proposed.First,the traditional guided filter was extended to quaternion field,and the minimized cost function of the guided filtering is modified.The real part of the linear output in the guided filtering is suppressed by introducing the parameter l,as a result,the image energy is concentrated into the imaginary part,and the guided filter based on quaternion was obtained.Then,the original color image was modeled in a quaternion matrix form,and the R,G and B components of the color image are respectively treated as the three imaginary parts of the pure quaternion.The image is filtered as a whole by the quaternion guide filtering.Furthermore,with the proposed quaternion guided filter,the source image is decomposed into a base image and a detail image,and the detail image is enhanced by the enhancement algorithm.Finally,the base image and the detail-enhanced image were reconstructed to obtain an enhanced image.In this paper,we propose two kinds of image detail enhancement methods,the first one is the self-adaptive detail enhancement transform.In the model of the self-adaptive transformation,the larger absolute value of the input quantity becomes,the larger of the output quantity' absolute value also changes,this feature is in line with the energy distribution of the details and noises in the image,,the energy of the noises is often relatively small,which can obtain relatively small amplification factor as a function of the input coefficient,but the energy of the details is just on the contrary,as a function input,it can get a larger amplification factor,so the function model can be applied to the image detail enhancement method in this paper.The second one is the detail Salient enhancement method.The saliency enhancement transforms the color image into the gray scale,then,summarizes the difference between the grayscale value and the other pixel values in the neighborhood,and the corresponding salient value is obtained.In this method,the histogram can be used to obtain a saliency map quickly and easily.The image saliency is employed to magnify the detail layer in order to enhance details in an image.Using saliency extraction,the contrast of image interesting regions would be enhanced.And the image is enhanced with linear detail enhancement.Finally,the linear enhancement result and Saliency enhancement result are combined to obtain the final salient detail enhancement image,which does not only highlight the interest region of the image,but also enhance the details of other regions in the image.Experimental results show that the proposed method can produce high-quality detail enhancement.It not only makes image's edges more prominent and textures clearer,but also keeps color fidelity.Comparing with the existing enhancement methods,our proposed method improves the visual quality significantly and the objective evaluating indicators are also greatly improved.In order to smooth the details of the color image and keep the edges of the image unambiguous,the guided filter and the lifting wavelet are effectively combined to propose a multi-scale guided filter.Firstly,the image is decomposed into a low frequency subband and a plurality of high frequency subbands by lifting wavelet.In the decomposition of the lifting wavelet,compared to high-frequency information,the low-frequency information as the main energy of the image,contains not only the basic part of the image,but also most of the image details.Then,in the reconstruction process of the lifting wavelet,the low-frequency information of each scale is smoothed by the guide filter and the edges are not blurred.Finally,the reconstructed image is processed by the guided filter once more to remove the residual details as soon as possible.The experimental results show that the multi-scale guided filter can smooth the image's details and maintain the edge integrity.In this paper,the multi-scale guided filtering is applied into the image haze removal using dark channel prior and the transmission map is refined.Compared with the He(Dark Channel Prior)algorithm,the proposed algorithm improves the visual effect of the image in whole,enhances the contrast and details of the dehazing image,can effectively restore the scene information and preserve the edge information of the scene.At the same time,the objective evaluation indexes are improved.In this paper,multi-scale guided filtering is also applied to flash/noflash denoising.The flash image is a sharp and noise free image with rich details,but that will destroy the natural lighting,it can be used as the guided image of the multi-scale guided filtering.And the noflash image is noisy,but maintains the natural lighting of the ambient image,so it can be treated as the input image.The paper combines the best properties of flash and no-flash images to reduce noise in the no-flash image,make details abundant and show a clear and natural image.Compared with the guided filtering,multi-scale guided filtering does not only effectively remove the noise,but also can enhance the details of the image,the most important is that the enhanced image is closer to the look of the real scene.As to the problem of edge blurring in the smoothness of the large-scale details,the paper proposes a new edge preserving filter.The filter makes full use of the information in the image and blends the estimated "depth" information of the image with traditional bilateral filtering to smooth images.First of all,with atmospheric scattering model and dark channel prior,this paper estimated transmission to obtain "depth" information.Although the depth is not a real depth value in a strict sense,it could reflect the level information of the images in a certain extent;Secondly,the paper employs the estimated depth information to construct a hard-edge constraint in order to constrain the effect of each pixel value to the neighborhood on constructing the bilateral filter weights.In the hard-edge constraint,a threshold parameterT_d is introduced.We compute the depth difference between the pixel point and the center point in the neighborhood.If the difference is larger than the threshold valueT_d,it indicates that the pixel point is not in the same depth layer as the center point,then its weight in the neighborhood is set to 0,otherwise,the weight will participate in weighting operation in the bilateral filter.In depth bilateral filtering,the threshold value is critical and has a large effect on the image processing.If the value T_d is large enough,then the hard-edge constraint will not work,and the depth bilateral filter will become a traditional bilateral filter.Conversely,if the value is small enough,the depth bilateral filter will not have some functions of edge-preserving and smoothing details.In this paper,the depth of the bilateral filter is applied to the image detail enhancement,image edge extraction and image texture removal and other fields.Experimental results show that the method doesn't only preserve the advantages of the bilateral filter,especially making edges clearly highlighted,but also produces halo-free edge-preserving smoothing by adjusting the threshold at the same time,the objective evaluating indicators have been greatly improved.