A Survey of Demosaicing: Issues and Challenges

Transcription

1 A Survey of Demosaicing: Issues and Challenges Er. Simarpreet Kaur and Dr. Vijay Kumar Banga Abstract A demosaicing is really a digital image method used to re-establish the full color image from partial color samples result from image sensor overlaid with a color filtration array. It is known as Color Filter Array (CFA) interpolation. Today most of the digital cameras make use of a single CCD image sensor with Bayer pattern having alternate arrangement of red, blue and green colour filters, A demosaicing technique is to be applied to extract a full color image.the ensuing mosaic of color samples is refined to make high definition color image.the overall objective of this paper is to address the different demosaicing techniques to create a full resolution color image, and identify elementary variations among competitive approaches, in terms of PSRN ratios and other issues. The review of almost all the research shown that the luminance components are interpolated earliest and the chrominance components are reconstructed predicated on luminance data. This paper ends up with suitable research gaps i.e. various challenges and issues of demosaicing techniques which are required to future research Keywords Bayer pattern, CFA, Demosaicing, Image Processing Light because the crest sensitivity of the Human Visual system lay in between the intermediate wavelength which justifies the extra green sampling [5]. Fig 1. Illustration of optical paths for multichip digital camera C I. INTRODUCTION OMMERCIALLY available digital color cameras are derived from single charge coupled device(ccd) array and record color data by utilizing three or more color filters, each sample acquiring just one sample of the color spectrum [1]. In an ideal condition, three separate sensor devices must be utilized to recapture the red, blue and green planes that will be too costly and mechanically difficult for real implementations as shown in figure 1. [2]. In this a beam splitters are used alongside the optical pathway for projecting the image on three separate sensors as shown and color images of three full-channel are obtained with the use of color filter in front of each sensor. So this problem is overcome by introducing a Color Filter Array (CFA) to fully capture a color image using only one sensor [3].Among many CFA patterns the Bayer pattern is one of the most popularly used color filter array pattern shown in figure 2.which features blue and red filters at alternating pixel locations on rectangular grids in the horizontal and vertical directions and green filters are arranged in the quincunx pattern at the rest of the locations [4]. There are two times as many green filters as red or blue ones. With this design pattern half (50%) of the image resolution is focused on predicting the green color band catering to the human eye s larger sensitivity towards green Er. Simarpreet Kaur is M Tech scholar of Amritsar College of Engineering and Technology, Amritsar, Punjab, India (sim.preet330@gmail.com). Dr. Vijay Kumar Banga is working as a Principal and Head of ECE department in Amritsar College of Engineering and Technology, Amritsar, Punjab, India, (vijaykumar.banga@gmail.com). A CFA should be situated between the lens and the sensors to construct a color image which consists of a mosaic selective filters by which each CCD element samples only one component out of the three color components Red, Green or Blue as shown in figure 3. Fig 2. Bayer Pattern CFA A CFA generally has one color filter element for each sensor and then estimate the missing two color components. This interpolation process is generally known as demosaicing as of the mosaic structure of the CFA Bayer pattern. Demosaicing is the most significant area of the image processing used in digital cameras. The disappointment of the used demosaicing process may weaken the whole image quality significantly. That s why it has been taken as an active research for several years. Even though there has been recent attempts to present comprehensive demosaicing methods most demosaicing alternatives in the literature are produced for the Bayer pattern. The most frequent approach for interpolation of missing pixels is by using the spatial invariant strategy such as for example bilinear or bicubic interpolation. But wherever there's an immediate change in the color

2 that may lead to the false color artifacts. The product quality could be enhanced by making the use of interpolation over color differences to make the most of connection between the color channels. Though the shortage of spatial adaptiveness would still bound the performance of interpolation. Fig 3. Single CCD Camera The efficiency of the interpolation method depends upon the usage of both the spectral and spatial correlations [6]. II. IMAGE DEMOSAICING The reconstruction of full color images from a CFA based detector necessitate a process of manipulating the values of any other color separations at each pixel.the methods of these types are commonly referred as color interpolation. The image below shown in figure 4 shows the output from the Bayer layer image sensor each pixel has only Red, Green or Blue components [7].The original image is shown along with demosaicied rebuilding. This reconstructed image will be accurate in uniform colored areas but it has a loss of resolution and has edge arti facts. Common Demosaicing Artifacts A reconstructed image from a CCD with a Bayer pattern CFA measures only 33% information of the original image. Therefore various artifacts occurs as a result of demosaicing [3]. Two common types of artifacts are Zippering and False coloring. 1. Zippring Or Blurring Artifacts Zippering and blurring effect is that when there is one side effect of CFA demosaicing, which occurs generally along edges. Therefore edge blurring occurs along the edges in an on/off pattern [8] Figure 6 shows demosaicied image with edge blurring effect. Figure (a) features the fence poles in the background of the image as well as along the stripes in the shirt of the person. Image (b) shows the zippering in the license plate along the upper edge of the bumper and along its three uncovered characters. Last image (c) presents the zippering in the truck along the edges within the headlights and along the upper edge of the grill. Fig 6. Presents zippering artifacts of CFA demosaicing. Fig 4. Bayer Filter Sample [7] So a digital camera is means to reconstruct a whole RGB image using all above information. The resulting image is like the shown image in figure False Color Artifacts A regular and unfortunate artifact of CFA demosaicing is fake coloring. That artifact on average manifests along edges wherever quick or unpleasant adjustments in color arise as a result of miss interpolation crosswise, rather than the length of, an edge.figure7 shows the false color demosaicied images. Image (a) features the car s window where an alternating blue red pattern occurs all along the horizontal edges. Figure (b) shows the false color in the ford s logo lettering with high frequency information. Image (c) displays the left edge of the windshield with alternating pattern of Red and Blue as well as red blue highlights on brighter portions of the mirror. Fig 5. Reconstructed Image [7] Fig 7. Images showing false color demosaicing artifacts. To remove the false coloring several methods are available. Adaptive color Filter Array demosaicing and some other algorithms remove false color after demosaicing [14][15][16].Smooth hue transition interpolation is

3 during the process of demosaicing [17][18][19] to prevent the false color from the final image. Applications Image demosaicing is used to reconstruct the full resolution color image from incomplete samples to provide image enhancement which is used for enhancing a quality of images. The image enhancement have wide area of applications inflight imaging, Satellite imaging, health check imaging, in modern Digital mobile phone camera, remote sensing, Image Enhancement techniques used in many areas such as forensics, Astrophotography, Fingerprint matching, etc. It includes contrast enhancement, edge sharpening, blur reduction, removing noise are just some of the techniques used to make the images bright. Images obtained from fingerprint recognition, safety measures videos analysis and indulgence scene investigations are enhanced to help in identification of culprits and protection of victims. III. LITERATURE SURVEY This section contains the literature from the study of various research papers. Fang et al. (2012) [46] has proposed an easy frequencydomain analysis approach for joint demosaicing and sub pixelbased down-sampling of single sensor Bayer images. From this, they integrated demosaicing into down-sampling by directly performing sub-pixel based down-sampling within the Bayer domain, which means that the computational complexity is reduced. Gang et al. (2012) [47] has presented an adaptive demosaicing algorithm by exploiting both the non-local similarity and your regional correlation in along with filter array image. First, just about the most flat nonlocal image patches are searched within the searching window devoted to the estimated pixel. 2nd, the plot, which can be regarded the absolute most just like the current plot, is selected among the absolute easiest nonlocal patches. Third, in line with the similar degree and your regional correlation degree, the obtained nonlocal image patches together with the current patch are adaptively chosen to estimate the missing color sample. Maalouf et al. (2012) [48] has discussed a bandlet-based demosaicing means for color images. We get a spatial multiplexing type of color if you wish to locate the luminance together with the chrominance aspects of the acquired image. Then, a luminance filter must be used to reconstruct the luminance component. Thereafter, using the thought of maximal gradient of multivalued images, an extension cord from the bandlet representation for the outcome of multivalued images has proposed. Finally, demosaicing is conducted by merging the luminance and every one of the chrominance components within the multivalued bandlet transform domain. Chung et al. (2012) [49] has presented an effective decision-based demosaicing algorithm for Bayer images. An enhanced edge-sensing calculate known as enhanced integrated gradient is capable of promoting more gradient data from several color strength and color big difference planes inside directional compatibility constraint. An adaptive green plane advancement which performs using the enhanced integrated gradient is generally proposed to further improve the performance from the algorithm. Jimmy Li et al. (2012) [50] has proposed an adaptive tactic to avoid inclusion these bad pixels within the interpolation process has proposed. Only defective pixels which are considerably more advanced than their surrounding neighbors will likely be deemed badly and you will be corrected. This is achieved by adaptively varying your order of interpolation so your period of interpolation is shorter in case a bad pixel is at closer on the pixel being interpolated. The bad pixels can be found via a median-based multi-shell filter structure. Maschal et al. (2012) [51] has proposed two new noreference quality assessment algorithms. These algorithms provide family members comparison of two demosaicing algorithms by measuring the two common artifacts, zippering and false coloring, throughout their output images. The pioneer algorithm, the benefit slope evaluate, tests the typical sharpness of each one of the three shade routes, therefore calculating the relative edge reconstruction accuracy of each and every demosaicing algorithm. The particular algorithm, the false color calculate, estimates deviations out of your established regular color huge difference image product and functions on green-red and green-blue color huge difference planes, therefore calculating the red and blue channel reconstruction of each and every demosaicing algorithm. We examine and rank popular demosaicing algorithm applying these new algorithms. Dongjae Lee et al. (2012) [52] has proposed two-layer color filter array, with a full resolution in G channel. 5 resolutions in R/B channel. To evaluate the performance, easy demosaic formulas are presented. A two-layer color filter array for any premium image and demosaic algorithms to interpolate the proposed two-layer color filter array. Though demosaic and color filter array design methods are already improved together, recent research results point out the performance saturation since sub-sampling of color channels is inevitable. To get rid of this limitation, multilayer color filter arrays are already developed so you can get several color data at single pixel position. Pekkucuksen et al. (2013) [53] has presented a super easy edge strength filter to interpolate the missing color values adaptively. As the filter is instantly applicable on the Bayer mosaic pattern, we debate that the exact idea may very well be extended along with other mosaic patterns and describe its application on the Lukac mosaic pattern. Clearly outperforms other available solutions with regards to CPSNR. We believe the edge oriented, directional approach could make a good choice for other CFA patterns as well. Glotzbach et al. (2001) [54] asserted that saving money image must be used to provide high-frequency

4 and reduce aliasing in the red and blue images. First, the red and blue images are interpolated which has a rectangular low pass filter in line with the rectangular sampling grid. This fills within the missing values within the grid, but allows aliasing distortions into your red and blue output images. These output images will also be missing the high-frequency components needed to make a sharp image. However, because saving money image is sampled at better pay, the high-frequency information can be taken from saving money image to raise a short interpolation from the red and blue images. A horizontal high pass filter in addition to a vertical high pass filter is utilized by saving money image. Provided the high-frequency information how the low sampling rate from the red and blue images cannot preserve. Ling Shao et al. (2014) [55] proposed a content adaptive demosaicing strategy utilizing structure analysis and correlation involving the red, green and blue planes. Those two aspects were chosen in the classification associated with a block of pixels to produced trained filters. The planned technique aims to reconstruct a first class demosaicied image originating from a Bayer pattern in any color filter array efficiently. Experimental results showed that the proposed strategy performs comparatively as higher end methods. Xin Li et al. (2008) [56] provides a systematic survey of published work their review efforts to handle essential issues to demosaicing and recognize fundamental differences among competitive approaches. Their findings recommend many current works belong to the class of consecutive demosaicing - i.e., luminance channel is interpolated first and then chrominance channels are reconstructed predicated on recovered luminance information. Our contrast is performed on two knowledge models: IMAX top quality photographs (more challenging) and Kodak Photo CD (popular choice). Although many active demosaicing calculations attain firstclass efficiency on the Kodak knowledge collection, their efficiency on the IMAX one (images with high-saturation edges and varying-hue) degrades extensively. Such remark suggests the importance of correctly approaching the issue of inequality between thought product and remark knowledge in demosaicing, which requires more research on dilemmas such as for example product validation, check knowledge choice and efficiency evaluation. Rastislav Lukac et at. (2004) [57] presented a normalized color-ratio model for color tiller array (CFA) interpolation schemes. The proposed normalized model enforces the underlying modeling assumption in both smooth and highfrequency image regions. The utilization of the proposed nodal, which represents a generalization of the conventional cola-ratio model, can significantly boost the performance of most well-known CFA interpolators, in terms of both objective and subjective image quality measures. Aditi Majumder at el (2007) [58] applied a greedy algorithm to an image in its native resolution without the requirement of any costly image segmentation operation. They posed the contrast enhancement as an optimization problem that raises an objective function that defines the local average contrast enhancement (ACE) an image subject to restrictions that get a handle on the comparison development by a simple parameter τ. We expand this approach to color photographs wherever color is preserved while improving only the luminance contrast. In addition, we range the parameter τ spatially over the image to reach spatially picky enhancement. Ultimately, we show that the ACE defined by the aim function may act as a full to assess the comparison development accomplished for different methods and different variables thereof. IV. DEMOSAICING METHODS The image enhancement techniques are divided into two broad categories: Spatial Domain Methods This operates directly on pixels. 1. For Luminance Channel Interpolation Earlier work the missing data in G channel is estimated or calculated by the heuristic edge-directed rules. The local edge direction is estimation from available [20]local 3 3 window or R/B [21] a local 5 5 window or both [22] where the second order gradients of chrominance channels can be used as correction terms. In primary- consistent soft decision (PCSD) scheme [23], G pixels are first cautiously interpolated alongside horizontal/vertical directions and then selected based on large framework. 2. For Chrominance Channel Interpolation Based on the interpolation full resolution G channel, the chrominance (R/B channels are then reconstructed by enforcing constant-shade(hue) rules. mainly, color-difference signals d RG = R G; d BG = B G are interpolated based on fullresolution G channel and down-sampled R/B channels; then G channel is simply added back to two color-difference channels for recovering R/B. In early works, color ratio [24] or colordifference [25] interpolation is frequently implemented by standard linear interpolation. Frequency Domain Methods This directly operates on Foriour Transform of an image. 1. For Luminance Channel Interpolation The deterministic formula that is =,, s =,, MsS Full-resolution image is converted to a mosaic observation according to the CFA sampling pattern. In this equation the sub sample color channels are R, G, B. and the front(mask) Ms takes a color sample at a pixel according to the CFA pattern. This formula suggests the feasibility of recovering missing data through frequency domain filtering. In an early frequency domain approach, [26] ad-hoc diamond-shape low pass (LP) filter is put on to reconstruct G channel. Later, a wavelet-based green channel upgrade method [27] is created predicted on stronger inter-channel correlation of highfrequency band

5 In the latest work, [28] another extension of frequency domain demosaicing [29] is offered where adaptive filtering (AF) is applied on the luminance (G) component. The luminance (G) criterion at green places are interpolated using a filter similar to the one in[29]; while the values at red/blue locations are estimated as a weighted sum of neighbouring luminance values, where the selection of weight is according to the horizontal and vertical gradients boundary. 2. For Chrominance Channel Interpolation Renovation of chrominance channels may also be executed by developing anti aliasing filters in the frequency domain (FD). In the very easiest condition, bilinear filters are accustomed to retrieve chrominance channels R L; G L; B L (L denotes the luminance channel) [29]. Improved linear filters for chrominance channels are produced [30], [31] to reduce the spectral crosstalk. Analogous linear interpolation algorithms may also be utilized in more recent frequencydomain methods [32], [28]. It s been generally observed that reconstruction errors in R/B channels are often significantly larger than those in G channel due to more severe aliasing. Yet another successful strategy of interpolating missing information in chrominance channels would be to borrow wavelet theory. On the basis of the observation that edges across color channels are extremely related, with the simple max rule [33] it can be possible to obtain a good estimation of high-frequency band coefficients in chrominance from their luminance counterpart V. PERFORMANCE EVALUATION OF DEMOSAICING ALGORITHMS A frequently applied quantitative evaluation for analyzing demosaicing algorithms is mean square error (MSE) or equivalently peak signal to noise ratio (PSNR). In some works, [34], [36] mean absolute difference (MAD) and normalized color difference (NCD) [35] may also be used as supplementary criterion. Dozens of target measures suppose the accessibility to a reference image. However, such prediction does not always hold since the output of any digital camera has already been processed by a pipeline including optical low-pass filtering and demosaicing. Because various cameras may have various demosaicing algorithms, some bias is inevitable, which deviate from the ground truth (an image acquired from 3CCD camera). To the best of our knowledge, computer-based simulation has been used in all published works on demosaicing - i.e., full-resolution color images acquired from single-ccd cameras are first down-sampled according to a specific CFA pattern and then the reconstructed images are compared with the original quantitatively. The only explanation we have for this approach is that the original full-resolution color images do visually appear pleasant (therefore can be used as the reference even if they have been through processing pipeline). VI. EXPERIMENTAL RESULT This section contains experimental result from comparative study among eleven selective inter-channel demosaicing algorithms on two different benchmark data sets. The ten demosaicing algorithms are: 1) Lu&Tan s method (LT)[37]; 2) alternating projection (AP)[27]; 3) adaptive homogeneitydirected (AHD)[32]; 4) successive approximation (SA) with edge-weighted improvement[38]; 5) Lukac s CCA method[36] with post-processing[15]; 6) Frequency-Domain (FD) demosaicing[39]; 7) Directional Filtering and a posterior Decision (DFPD)[40]; 8) Variance of color-difference (VCD)[41]; 9) Directional Linear Minimum Mean Square- Error Estimation (DLMMSE)[42]; 10) local polynomial approximation (LPA)[43]; 11) Adaptive filtering (AF) for demosaicing.[44]. The two bench mark data set is from IMAX and Kodak photo CD high quality images. The PSNRs are calculated as the objective measures for comparing the demosiacing algorithms. As various operations skips the processing of pixels at the border, we exclude those pixels whose distance to the border is fewer than 10 pixels in calculation.. Due to the variation of implementations (e.g., MATLAB vs. C codes), the actual running time does not faithfully reflect the computational complexity of each algorithm. Therefore, we choose not to report complexity-related measures Table I contains the performance contrast on IMAX images data set. The hue saturation characteristics of IMAX images are arguably closer to those of images acquired by digital cameras of these days. For this data set, we have tested all selected 11 algorithms with one replacement: FD is replaced by new edge directed interpolation NEDI [45] (an intrachannel interpolation technique) since AF represents a more competing frequency domain approach. The rationale behind that is that we want to include Ή intra into our comparison. It can be observed from the table that the performance of the most existing algorithms degrade a lot on this more challenging data set. Many inter-channel demosaicing do not necessarily outperform NEDI (intra-channel method). It should be noted that the algorithm [37] has shown impressive performance on this new data set, which suggests the importance of jointly exploiting spatial and spectral correlation and effectiveness of weighted interpolation Despite the popularity of Kodak Photo CD images, they are relatively low-quality representation of natural world (they are scanned version from film-based photos).table II includes the performance comparison on the Kodak Photo CD data set. The above two experiments suggest that the issue of mismatch between assumed model and observation data has to be appropriately addressed. One way out to reduce the risk of mismatch is to fuse the demosaicing results by different

6 Fig 8. Image set from Kodak Photo CD (left) and cropped image set from IMAX (right) TABLE I PERFORMANCE COMPARISON OF DIFFERENT DEMOSAICING METHODS FOR PSNR (DB) ON IMAGE SET FOR CROPPED IMAX No. AP SA AF DFPD VCD DL NEDI CCA AHD LT LPA ` Avg TABLE II COMPARISON PERFORMANCE OF DIFFERENT DEMOSAICING METHODS FOR PSNR (DB) ON IMAGE SET OF CROPPED KODAK PHOTO AS DL IS SHORT FORM FOR DLMMSE No. AP SA AF DFPD VCD DL FD CCA AHD LT

7 Avg VII. GAPS IN LITERATURE WORK It has been found that most of the existing literature does not focus on at least one of the following things: Color Artifacts The majority of the image enhancement methods are transform domain so may come up with certain artifacts in output image so require some special aid to overcome illuminate artifacts. Pixel Lost As a result of transform domain methods certain pixels may get lost during conversion either original to transform or transformed signal to original pixel values Uneven Illuminate The problem of uneven illuminate has been ignored in the majority of existing research on color filter array. Most of the methods depends upon certain predefined rules no concentrate on the objects or regions in the given image; so may imbalance the illuminate of the output image. VIII. CONCLUSION AND FUTURE SCOPE In this survey paper we have reviewed various image demosiacing techniques that show the luminance channel is recovered first and then based on full resolution luminance image, chrominance channels are reconstructed. This survey demonstrated the importance of jointly exploiting spatial and spectral correlations especially for images with high resolution and varying hue. The directions required for the future work are first, the demosaicing of images with weak spectral correlation remains a challenging task. Second, the performance evaluation of demosaicing algorithms needs more careful investigations as if the images acquired by only single CCD cameras more cautions needed about the risk with computers based simulation. Moreover to reduce the color artifacts illuminate normalization technique based on gray world constancy can be used to balance the colored

8 REFERENCES [1] Ramanath, Rajeev, Wesley E. Snyder, Griff L. Bilbro and William A. Sander et al. "Demosaicing methods for Bayer color arrays."journal of Electronic imaging 11.3 (2002): [2] Shao, Ling, and Amin Ur Rehman. "Image demosaicing using content and color-correlation analysis." Signal Processing 103 (2014): [3] Maschal, Robert A., et al. "New image quality assessment algorithms for CFA demosaicing." Sensors Journal, IEEE 13.1 (2013): [4] B. E. Bayer, Color image array, U.S. Patent , Jul. 15, [5] X. Li, B. Gunturk, and L. Zhang, Image demosaicing: A systematic survey, Proc. SPIE, Visual Commun. Image Process. vol. 6822, pp.68221j J-15, Jan [6] Mohanbabbu, vinothkumar and ponvasanth, et al. A Modified Low Power Color Filter Array. Interpolation International Conference on Communication and Signal Processing, April3-5,2014,India [7] R. Lukac and K. N. Plataniotis, A normalized model for color-ratio based demosaicing schemes, in Int. Conf. on Image Process., 2004, vol. 3, pp [8] W. Lu and Y.-P. Tan, Color filter array demosaicing: New method and performance measures, IEEE Trans. Image Process., vol. 12, no. 10, pp , Oct [9] J. E. Adams and J. F. Hamilton, Jr., Adaptive color plane interpolation in single sensor color electronic camera, U.S. Patent , Jul. 29, [10] H.-A. Chang and H. Chen, Directionally weighted color interpolation for digital cameras, in Proc. IEEE Int. Symp. Circuits Syst., vol. 6. May 2005, pp [11] R. K. Thakur, A. Tripathy, and A. K. Ray, A new edge adaptive and directionally weighted color interpolation algorithm of digital camera images, in Proc. Int. Conf. Adv. Comput., Commun. Control, Jan. 2009, pp [12] D. R. Cok, Signal processing method and apparatus for sampled image signals, U.S. Patent , Dec.16, [13] C. Popescu, Statistical tools for digital image forensics, Ph.D. dissertation Dept. Comput. Sci., Dartmouth College, Hanover, NH, Dec. 2004, pp [14] L. Chang and Y.-P. Tan, Adaptive color filter array demosaicing with artifact suppression, in Proc. IEEE Int. Symp. Circuits Syst., vol. 3. May 2004, pp [15] R. Lukac, K. Martin, and K. N. Plataniotis, Demosaicied image post processing using local color ratios, IEEE Trans. Circuits Syst. Video Technol., vol. 14, no. 6, pp , Jun [16] W. T. Freeman, Median filter for reconstructing missing color samples, U.S. Patent , Feb. 9, [17] J.E. Adams, Interactions between color plane interpolation and other image processing functions in electronic photography, Proc. SPIE, Cameras Syst. Electron. Photography Sci. Image., vol. 2416, pp , Mar [18] R.Ramanath, W. E. Snyder, G. L. Bilbro, and W. A. Sander, III, Demosaicing methods for Bayer color arrays, J. Electron. Image. vol. 11, no. 3, pp , Jul [19] R.Lukac and K. N. Platanioutis, A normalized model for color-ratio based demosaicing schemes, in Proc. Int. Conf. Image Process., 2004, pp [20] Robert H. Hibbard. Apparatus and method for adaptively interpolating a full color image utilizing luminance gradients. U.S. Patent , Jan [21] C. Laroche and M. Prescott. Apparatus and method for adaptively interpolating a full color image utilizing chrominance gradients. U.S. Patent , Dec [22] J. F. Hamilton Jr. and J. E. Adams. Adaptive color plane interpolation in single color electronic camera.u.s. Patent , May [23] X. Wu and N. Zhung. Primary-consistent soft-decision color demosaicing for digital cameras. IEEE Trans. on IP, 13(9): , September 2004 [24] D.R. Cok. Signal processing method and apparatus for producing interpolated chrominance values in a sampled color image signal. U.S. Patent , Feb [25] S.C. Pei and I.K. Tam. Effective color interpolation in CCD color filter array using signal correlation. CirSysVideo, 13(6): , June [26] J. W. Glotzbach, R. W. Schafer, and K. Illgner. A method of color filter array interpolation with alias cancellation properties. In Proc. IEEE Int. Conf. Image Processing, volume 1, pages , 2001 [27] B. Gunturk, Y. Altunbasak, and R. M. Mersereau. Color plane interpolation using alternating projections. IEEE Transactions on Image Processing, 11(9): , 2002 [28] N.-X. Lian, L. Chang, Y.-P. Tan and V. Zagorodnov. Adaptive filtering for color filter array demosaicing. IEEE Transactions on Image Processing, 16(10): , [29] D. Alleysson, S. Susstrunk, and J. Herault. Linear demosaicing inspired by the human visual system. IEEE Trans. on IP, 14(4): , April [30] E. Dubois. Frequency-domain methods for demosaicing of Bayersampled color images. IEEE Signal Proc. Letters, 12(12): , December [31] E. Dubois. Filter design for adaptive frequency-domain Bayer demosaicing. In Proc. of International Conference on Image Processing, pages , [32] K.Hirakawa and T.W. Parks. Adaptive homogeneity-directed demosaicing algorithm. IEEE Trans. on IP, 14(3): , March2005. [33] J. Driesen and P. Scheunders. Wavelet-based color filter array demosaicing. In Proc. ICIP, pages V: , [34] R. Lukac, K. Martin, and K.N.Plataniotis. Demosaicied image post processing using local color ratios. IEEE Trans. Cir. Sys. Video Tech., 14(6): , June [35] Konstantinos N. Plataniotis and Anastasios N. Venetsanopoulos. Color image processing and applications. Springer-Verlag New York, Inc., New York, NY, USA, [36] R. Lukac, K.N. Plataniotis, D. Hatzinakos, and M. Aleksix. A novel cost effective demosaicing approach. IEEE Transactions on Consumer Electronics, 50(1): , [37] W. Lu and Y.P. Tan. Color filter array demosaicing: New method and performance measures. IEEE Trans. on IP, 12(10): , October [38] Chung-Yen Su. Highly effective iterative demosaicing using weightededge and color-difference interpolations. IEEE Transactions on Consumer Electronics, 52(2): , May [39] D. Alleysson, S. Susstrunk, and J. Herault. Linear demosaicing inspired by the human visual system. IEEE Trans. on IP, 14(4): , April [40] D. Menon, S. Andriani, and G. Calvagno. Demosaicing with directional filtering and a posteriori decision. IEEE Trans. on IP, 16(1): , Jan [41] K.-H. Chung and Y.-H. Chan. Color demosaicing using variance of color differences. IEEE Trans. on IP, 15(10): , Oct [42] L. Zhang and X. Wu. Color demosaicing via directional linear minimum mean square-error estimation. IEEE Trans. on IP, 14(12): , December 2005 [43] Dmitriy Paliy, Vladimir Katkovnik, Radu Bilcu, Sakari Alenius, and Karen Egiazarian. Spatially adaptive color filter array interpolation for noiseless and noisy data. International Journal of Imaging Systems and Technology, 17(3): , Oct [44] N.-X. Lian, L. Chang, Y.-P. Tan and V. Zagorodnov. Adaptive filtering for color filter array demosaicing. IEEE Transactions on Image Processing, 16(10): , [45] X. Li and M. Orchard. New edge directed interpolation. IEEE Trans. on Image Proc., 10: ,

9 [46] Fang, Lu, Oscar C. Au, Yan Chen, Aggelos K. Katsaggelos, Hanli Wang, and Xing Wen. "Joint Demosaicing and Sub pixel-based Down- Sampling for Bayer Images: A Fast Frequency-Domain Analysis Approach." Multimedia, IEEE Transactions on 14, no. 4 (2012): [47] Guo-gang, Wang, Zhu Xiu-chang, and GanZong-liang. "Image demosaicing by non-local similarity and local correlation." In Signal Processing (ICSP), 2012 IEEE 11th International Conference on, 2, pp IEEE, 2012 [48] Maalouf, Aldo, M-C. Larabi and S. Susstrunk. "An efficient demosaicing technique using geometrical information." In Image Processing (ICIP), 2012 IEEE 19th international Conference on, pp IEEE, [49] Chung, King-Hong, and Yuk-Hee Chan. "Enhanced integrated gradient and its applications to color demosaicing." In Signal Processing, Communication and Computing (ICSPCC), 2012 Signal Processing, Communication and Computing (ICSPCC), 2012 IEEE International Conference on, pp IEEE, [50] Li, Jim S. Jimmy, and Sharmil Randhawa. "Integration of Adaptive CFA Demosaicing with Bad Pixel Removal." In Pattern Recognition, pp Springer Berlin Heidelberg, [51] Maschal, R. A., S. Susan Young, Joseph P. Reynolds, Keith Krapels, Jonathan Fanning, and Ted Corbin. "New image quality assessment algorithms for CFA demosaicing." Sensors Journal, IEEE 13, no. 1 (2013): [52] Lee, Dongjae, Byung-JoonBaek, and Tae-Chan Kim. "Two-layer color filter array for high quality images." In Image Processing (ICIP), th IEEE International Conference on, pp IEEE, [53] Pekkucuksen, Ibrahim, and Yucel Altunbasak. "Multiscale Gradients Based Color Filter Array Interpolation." (2013): 1-1. [54] Glotzbach, John W., Ronald W. Schafer, and Klaus Illgner. "A method of color filter array interpolation with alias cancellation properties." In Image Processing, Proceedings International Conference on, vol. 1, pp IEEE, [55] Shao, Ling, and Amin Ur Rehman. "Image demosaicing using content and color- correlation analysis." Signal Processing 103 (2014): [56] DIP [LAST VISITED] 29-JULY-2013 [ONLINE AVAILABLE] 2yYmT2RjyXU1gIa5Pf1FfRk3IaaWz9gUvzkQd. [57] R. Lukac and K. N. Plataniotis, A normalized model for color-ratio based demosaicing schemes, in Int. Conf. on Image Process., 2004, vol. 3, pp