A Robust Image Restoration by Using Dark channel Removal Method

Transcription

1 Volume 6, Issue 3, Marh 2017, ISSN: A Robust Image Restoration by Using Dark hannel Removal Method Ankit Jain 1 (MTeh. sholar), Prof. Mahima Jain 2 Department Of Computer Siene And Engineering, Bansal Institute of siene and tehnology, Bhopal, , India Abstrat This work fous on Fog removal whih is also known as visibility restoration refers to different methods that aim to redue or remove the degradation. The degradation may be due to various fators like relative, relative atmospheri turbulene, objet-amera motion, miss-fous, blur due to amera and others. This paper has utilized the dark hannel method for estimating the atmospheri light with hange in environment. This help in removing the distortion present in the image. Experiment was done on real dataset as well as on artifiial dataset images. It has been obtained that that proposed work has improve edge restoration value by 2.4 time, while ontrast restoration value was inreased by 4.51 time as ompare to previous work is better as ompare to previous work in [8]. Index Terms Digital Image Proessing, Haze, Information Extration, Fog removal. Visibility restoration I. INTRODUCTION Nowadays digital ameras are ertainly the most used devies to apture images. As ameras are present in very important user eletroni devie that is mobile phones, palmtop, omputer, surveillane system, robot et. As it an be seen or ompare from the previous amera images image quality is inreasing day by day with inrease in eletroni media. This is possible beause of the availability of large memory for proessing with high speed proessor in the same devie. As all kind of images taken from the digital amera have one ommon transparent medium that is air. As it was assumed that light rays reflet from the objets of the sene and get bak to the observer without any hanges or alterations. So in this ondition light in the image or brightness of the image is dependent on the single point of the sene. So all the algorithms and digital iruit with sensors are designed to work under lear light. So vision system is highly required to work under unlear atmospheri onditions suh as fog, rain, snow et. As anient painting and painters shows there painting with atmospheri onditions of nature. In their work optial phenomena was present in form of blue haze from different mountains and visibility of the objets got redued beause of bad nature onditions. As Leonardo da Vini painting ontain the atmospheri baground sene with few points are bluer and brighter [10]. As this optial phenomena is an attrative point for the images but in the omputer vision system it at as an noise for identifying the objets. As in order to inrease the seurity of the different important plaes some of the outdoor surveillane ameras are used for inreasing the robustness. But some time due to bad weather ondition it is seen that images or video obtain from the amera was not lear and there is no mehanism to improve the visibility through the ameras. /This is beause of the drasti hange in the ontrast and olor of the reeiving light from the image objets. So by using normal image proessing algorithms this images of bad weather ondition annot be improved for getting suffiient information. 208

2 Volume 6, Issue 3, Marh 2017, ISSN: Haze removal is a tough task beause fog depends on the unknown sene depth information. Fog effet is the funtion of distane between amera and objet. Hene removal of fog requires the estimation of air light map or depth map. The urrent haze removal method an be divided into two ategories: image enhanement and image restoration. Image enhanement does not inlude the reasons of fog degrading image quality. This method an improve the ontrast of haze image but loses some of the information regarding image. Image restoration firstly studies the physial proess of image imaging in foggy weather. After observing that degradation model of fog image will be established. At last, the degradation proess is inverted to generate the fog free image without the degradation. So, the quality of degraded image ould be improved. II. RELATED WORK Z. Zhu et. al. [1] introdued an experiene fusion method for various images by way of moving objets. The proposed method onsist a ghost removal algorithm in a low dynami series domain and a exposure fusion algorithm. The proposed ghost removal algorithm inludes a bidiretional normalization-based method for the finding of non-reliable pixels and a two-round hybrid method for the orretion of non-onstant pixels. A exposure fusion algorithm onsist a ontent adaptive bilateral filter,that extrats superior details from all the orreted images onurrently in asent domain. The final image is synthesized by seletively adding the extrated fine details to an in-between image that is generated by fusing all the orreted images via an existing multi-level algorithm. experiments and find that, if dealing with a haze image with large bakground area and low ontrast, dark hannel prior result will beome dark, also a general haze image after dark hannel ours different degree of anamorphous. They introdued an adaptive algorithm to repair the different kinds of an amorphose on the hazy image after dark hannel prior. Khairi Abdulrahim et. al. [3] projeted a fast yet tough tehnique to enhane the visibility of video frames using the dark hannel prior united with fuzzy logi-based tehnique. The dark hannel prior is a arithmetial uniformity of outdoor haze-free images based on the examination that most loal pathes in the haze-free images have pixels whih are dark in at least one olor hannel, where the fuzzy logi-based tehnique is used to plan an input spae to an output spae using a olletion of fuzzy membership funtions and poliy to deide deliately in ase of doubts. The ombination of the dark hannel and the fuzzy logi-based tehnique will make high quality haze-free images in real-time. Z. Bahok. Et. al. [5] has disussed that the within the last deades, improving the quality of an underwater image has one problem that is poor visibility of the image whih is aroused by physial properties of the water medium. A.C. Bovik et. al. [6] proposed pereptual models that an be able to foreast the value of distorted images with as little prior information of the images or their deformation as possible. The new IQA model, whih is known as Natural Image Quality Evaluator is based on the prodution of a quality aware olletion of statistial features based on a simple and suessful spae area natural sene statisti model. Canmei Yang et. al. [2] desribed a novel and effiient single image enhanement algorithm for haze image. As they monitor that, the ontrast and intensity of haze image after using dark hannel prior approah will neessarily tend to be lower than those of the real sene, they used histogram requirement to make an enhanement on image after dark hannel prior approah. They made a large number of X. Tang et. al. [7] proposed novel widespread guided image filtering method with the suggestion image generated by signal sub-spae projetion tehnique. It aepts ompliated parallel study through Monte Carlo imitation to hoose the dimensions of signal subspae in the path-based noisy images. The noise free image is reonstruted from the noisy image expeted onto the signifiant images by omponent 209

3 Volume 6, Issue 3, Marh 2017, ISSN: analysis. Test images are utilized to deide the relationship between the most favorable parameter value and noise divergene that maximizes the output peak signal-to-noise ratio. In [8] has presented a new method alled mixture CLAHE olor models that speifially developed for underwater image enhanement. The proess performs CLAHE method on RGB and HSV olor models. The projeted tehnique has onsiderably enhaned the visual superiority of underwater digital images by enhaning illuminate, as well as dropping noise and artifats. as "box" or "ross" patterns). Note that if the window has an odd number of entries, then the median is simple to define: it is just the middle value after all the entries in the window are sorted numerially. Image Dataset Pre-Proessing III. PROPOSED METHODOLOGY This paper fous on the digital hazy image restoration. Here image store the edge region of the image then apply Laplae distribution for pixel value restoration. Here whole work is explained in fig. 1. Pre-Proessing Here as the image is the olletion of pixels where eah pixel is representing a number that is refleting a number over there now for eah number depend on the format it has its range. So read a image means making a matrix of the same dimension of the image then fill the matrix orrespond to the pixel value of the image at the ell in the matrix. Median Filter 16X16 Blok De- Fusion For Eah Blok Median Filter The main idea of the median filter is to run through the signal entry by entry, replaing eah entry with the median of neighboring entries. The pattern of neighbors is alled the "window", whih slides, entry by entry, over the entire signal. For 1D signals, the most obvious window is just the first few preeding and following entries, whereas for 2D (or higher-dimensional) signals suh as images, more omplex window patterns are possible (suh Adjustable Color Parameters Haze Thikness Estimation Combine Blok Fig. 1. Blok diagram of proposed Restoration Image. 210

4 Volume 6, Issue 3, Marh 2017, ISSN: X16 Blok: As work is done on olor image so embedding is done on the red matrix of the image, so whole operation of embedding is done this red matrix. Whole red, green, blue matrix is divide into 16X16 bloks for restoration of image. In above equation (2) S is the region in the blok of image having hannel. C represent olor band Red, Green, Blue, et. the effet of fog is Air light effet and it is alulate as: AirLight(x) = A(1 t(x)) (3) Dark hannel prior: In [2] Dark hannel tehnique is developed in order to alulate the atmospheri light in the image. So it is emerged as a ommon tehnique in non sky part of the image beause few olor hannels has very less intensity in the few pixels. Here in dark olor hannel low intensity is present beause of the below three omponents: i). Surfae Colourful objets suh as grass, trees, et. In this J is olor image omprising of RGB omponents, represents a loal path whih has its origin at x. The low intensity of dark hannels is attributed mainly due to shadows in images, saturated olor objets and dark objets in images. After dark hannel prior, one need to estimate transmission t(x) for proeeding further with the solution. Another assumption needed is that let Atmospheri light A is also known. This normalize (4) by dividing both sides by A: ii). Shadow of tree, building, pillers, et. iii). Any high intensity objet surfae suh as blak stone, trunk, et. J ( I ( x) A (1 t( x))) ( x) (4) max( t( x), t ) o So most of outdoor image is full of above three points whih inlude olorful objet, few shadows and dark hannels whih fill image with noise. In presene of fog in environment image get brighter then atual image without fog. So it an be onlude that dark hannel of the image have high intensity of image in region with higher haze. So in order to find the light intensity an approx value is find by estimating the thikness of the haze. In ase of shady hannel prior this tehnique use pre and post proessing steps in order to improve results. In post proessing stepladder tehnique use flexible matting or twosided filtering et. This an be understand as if J(x) is input image, I(x) is hazy image, t(x) is the transmission of the environment. The redution of image beause of presene of fog an be alulate by: Here A is atmospheri light adjustment parameter in the blok so for eah blok it is evaluate by A min(max( S) (1) I t( x) min A (2) Figure 2: Haze removal results. Top: input haze images. Middle: restored haze-free images. Bottom: depth maps. 211

5 Volume 6, Issue 3, Marh 2017, ISSN: IV. EXPERIMENT AND RESULT In this setion, first introdue experimental settings, and then present the experimental results that validate the effetiveness of the approah. The experiments atually ontain two parts. This work is ompare with other previous work in [8] whih have utilize the laplae and haze thikness estimation only. Evaluation Parameters Visible edge restoration parameter: The e metri represents the rate of visible edge restoration in the haze-free image and is given by e V r V V o o (5) where Vr and Vo represent the total number of visible edges within the restored hazy image and the inoming hazy image, respetively. Contrast restoration: r metri is used to express the quality of ontrast restoration within the haze-free image. As suh, the.r metri is formulated as follows: 1 r exp Vr p i i log( ri ) (6) where Pi is the orresponding element within the set r, and ri is the rate of gradients between the restored hazy image and the inoming hazy image. Note that r onsists of the visible edges in the restored hazy image. Over or Under Exposed Metri: Moreover, the σ metri represents the number of pixels that might be either overexposed as white or underexposed as blak in the restored image. The σ metri is alulated as follows: dim V s x dim y where Vs represents the total of both overexposed and underexposed pixels in the restored image, and dimx dimy represents the size of the inoming image. Results Images Visible Edge Restoration Proposed Previous 1 Fog Sand Water Table 1. Comparison of proposed work and previous work on visible edge restoration parameter. In table 1 It is obtained that proposed work is better as ompare to previous as edge restoration value of proposed work is higher as ompare to previous. So inlusion of edge feature in haze removal has inrease the performane of the work. Images Contrast Restoration Image Proposed Previous 1 Fog Sand Water Table 2. Comparison of proposed work and previous work on visible edge restoration parameter. In table 2 It is obtained that proposed work is better as ompare to previous as ontrast restoration image value of proposed work is higher as ompare to previous. So inlusion of edge feature in haze removal has inrease the performane of the work. 212

6 Volume 6, Issue 3, Marh 2017, ISSN: Images Over or Under Exposed Metri Proposed Previous 1 Fog Sand Water Table 3. Comparison of proposed work and previous work on visible edge restoration parameter. In table 3 It is obtained that proposed work is better as ompare to previous as Over or Under Exposed Metri of Restoration image value of proposed work is higher as ompare to previous. So inlusion of edge feature in haze removal has inrease the performane of the work. Proessing, Communiations And Computing (Isp), 2014 Ieee InternationalConferene On, Pp Ieee, [3] Alajarmeh, Ahmad, Rosalina Abdul Salam, Mohd Fadzli Marhusin, And Khairi Abdulrahim. "Realtime Video Enhanement For Various Weather Conditions Using Dark Channel And Fuzzy Logi." In Computer And Information Sienes (Ioins), 2014 International Conferene On, Pp Ieee, [4] Serikawa, Seiihi, And Huimin Lu. "Underwater Image Dehazing Using Joint Trilateral Filter." Computers & Eletrial Engineering 40.1 (2014): [5] Hitam, M. S., W. N. J. H. W. Yussof, E. A. Awalludin, And Z. Bahok. "Mixture Contrast Limited Adaptive Histogram Equalization For Underwater Image Enhanement." In Computer Appliations Tehnology (Iat), 2013 International Conferene On, Pp Ieee, V. CONCLUSIONS A new ombination of median filter with dark hannel method is used in this work for dehazing image from different sene. The algorithm removes spatially varying haze based on the haze thikness estimation. As experiment is done on images of different environment and it is obtained that proposed work is better on all the evaluation parameters of de-hazing images. In Future improvements of the method will deal with possible orner, and histogram effets aused by the image proessing. REFERENCES [1] Z. Li, J. Zheng, And Z. Zhu, Contenet Adaptive Guided Image Filtering. In Ieee Int. Conf. Multimedia And Expo (Ime), 2014, [2] Liu, Feng, And Canmei Yang. "A Fast Method For Single Image Dehazing Using Dark Channel Prior." In Signal [6] A.Mittal, R. Soundararajan, And A.C. Bovik, Making A Completely Blind Image Quality Analyzer, Proess. Lett., Vol. 20, No.3, Pp , Mar Ieee Signal [7] K. He, J. Sun, And X. Tang, Guided Image Filtering, Ieee Trans. Patt. Anal. Mah. Intell., Vol. 35, No. 6, Pp ,2013. [8] Zhu, Qingsong, Shuai Yang, Pheng Ann Heng, And Xuelong Li. "An Adaptive And Effetive Single Image Dehazing Algorithm Based On Dark Channel Prior." Inrobotis And Biomimetis (Robio),2013 Ieee International Conferene On, Pp Ieee, Ankit Jain, MTeh. Sholar, Department Of Computer Siene And Engineering, Bansal Institute of siene and tehnology, Bhopal, , India Mahima Jain, Prof. in the Department Of Computer Siene And Engineering, Bansal Institute of siene and tehnology, Bhopal, , India 213