Efficient Color Image Enhancement Using Computational Photography Method

Shailendra Singh Rajpoot

M. Tech. Student, Department of Electronic and Communication Engineering, VNS Faculty of Engineering, Bhopal (Madhya Pradesh), INDIA

* Correspondence: E-mail: ssrajpoot1989@gmail.com

(Received 21 July, 2017; Accepted 24 Aug, 2018; Published 06 Sept, 2018)

ABSTRACT: In this letter, we propose a new global contrast enhancement algorithm using the computational photography method. On the basis of the computational photography, the color and depth image are ?rst partitioned into subintervals using the Gaussian mixture model. The positions partitioning the color are then adjusted such that spatially neighbouring pixels with the similar intensity and depth values can be grouped into the same sub-interval. By approximating the mapping curve of the contrast enhancement for each sub-interval, the comprehensive image contrast can be improved without over-enhancing the local image contrast. Experimental results determine the effectiveness of the proposed algorithm.

Keywords: Computational photography; color; depth image; Gaussian mixture model; mapping curve and contrast enhancement.

INTRODUCTION: It is observed that the image enhancement is a crucial pre-processing step for many computer vision systems. Sometime, the acquisition devices fail to adapt the high dynamics changes in the scene. So many image enhancement processes has been proposed in the literature that apply on different sets of criteria. As seen, the scene or image may be affected due to many reason such as change in illumination, shadows cast by neighbour object, fail in range handling capacity of device, noise during acquisition, transmission and reception. The above factors may affect and change the actual colour, brightness, contrast. Therefore, the multiscale image enhancement has become the great area of research in computer vision. The challenges before one has to preserve the color, edges, tonal and proper contrast of the images. However, a direct multi-scale image enhancement algorithm capable of independently and/or simultaneously providing adequate contrast enhancement, tonal rendition, dynamic range compression, and accurate edge preservation in a controlled manner has yet to be produced. 19

The improvement in contrast effectively recovers the visual quality that helps to understand the image content and distinguish the object in the region of interest from the background. The simplicity and quickness make the histogram equalization technique as most extensively used tool for image enhancement. Further to improve the local region of the image, the sub-image enhancement technique is proposed in the previous work. A non-linear diffusion equation is proposed to reduce the sudden illumination changes.1 The diffusion strength in textural image is estimated and the neighbour’s suppression is done for the degraded images. In this work, we have proposed the image enhancement techniques that utilize the contrast enhancement in the degraded region and preserve the colour of the input image. The other section of the paper is organized as follows. In Section II, literature review is done of some widely used effective image enhancement techniques. Proposed method is explained in Section III. The .Experimental results on some standard dataset are presented in Section IV. The paper is concluded in V.

LITERATURE REVIEW: Non-linear diffusion equation1 is utilized to improve the diffusive strength in the textural areas of an image. It achieved two benefits that the method is capable to preserve the boundary of the image due to the sudden illumination and also the halo artefact is eliminated. The second strength of the algorithm is to preserve the texture details of the images under the illumination condition. Since the illumination suppression requires the suppression of the pixels in images that may change the contrast and destroy the original information of the scene.

Optimization-based framework2 that employ the convolution term, fidelity and a prior term that regularizes the pixel value and obtained the image that is resembled with the original one. It utilizes the approximation of the median filtering process with the generalized Gaussian as the distribution model and estimated the pixels of the original one.

The GMM framework is used in this method that accumulates the similar patches using the multivariate Gaussian probability. It improves the local contrast greatly and preserves the tonal value of the images. The idea is patch based clustering approach that provides better goodness-of-fit to statistical properties of natural images. 3

Authors proposed used to preserve the backlight-scaled images as much as possible. It utilizes the luminance and chrominance components which account into an integral manner.4

In this method the entropy maximization process is used for the tone preserving. It constructed the K-edges maximum-weight path that optimizes the correct brightness and estimated the correct pixels at the spot.5

Weighted transform functions6 has been used to enhance the contrast of the image. The mean value is used to calculate the similarity and dissimilarity that further calculated the weighted transformation functions. The bin of the histogram is filtered out to increase the dynamic range of the enhanced image.

Real time filtering7 is proposed to enhance the finger print image. It split a modified anisotropic Gaussian filter into two orthogonal Gaussians and an oriented line Gaussian which in turns developed the architecture to adjust the dynamics of the scene.

Output image8 of low resolution camera and the high resolution RGN camera for exploiting the statistical correlation. It used the guided weight function for the dependency modelling that updated the depth of the image with a optimal restoration.

Histogram modification framework9 to enhance the colour and depth of the images. It partitioned the image into subinterval using the Gaussian mixture model. The spatially similar pixels having the same intensity level is grouped together. A mapping is proposed to enhance the depth and contrast of the image without over enhancing the contrast of the image.

Adaptive contrast enhancement algorithm10 by preserving the one dimensional histogram and the histogram obtained by gray level difference between the two neighboring pixels is proposed. The one dimensional histogram is utilized in enhancing the contrast of the image, while 2d histogram used to improve the detail of the frequently occurring the non smoothing area in an image.

MATERIALS AND METHODS : As seen, the limited dynamic ranges of imaging and display device limits the capability of handling the high dynamic range scenes. In the first stage, we have compressed the dynamic range and enhanced the local contrast using.12 It is focused to keep constant the tonality of image while increasing the luminance in the shadows region. In aid to this, local contrast enhancement method is applied to the above procedure. The above process may improve the quality of image without creating unnatural rendition in it. It utilizes the hyperbolic tangent part of the image that enhances the dark region of the image while preserve the light part region.

img2

img2

As we know that the HSV represents hue, saturation and value (brightness). The computational photography is applied to improve the quality of images by increasing the brightness without changing the colour component.12 If the input image contains lot of color, the value of N is taken as one while for the higher values of N, the transformed image will contain more contrast and strange effects. Since the alone v transform do not handle the contrast in local region and the adjusted the brightness globally. However the local contrast and intensity calculation is trade-off between the two parameters. The output image found to be change in color if we adjust the parameter to improve the contrast locally. In order to improve the contrast by maintaining the brightness in the dark region is proposed in the work. The reconstructed image is well adjusted near the dark region and the contrast is also improved by averaging the global intensity. The flow diagram of proposed method is shown in below figure 1.

img2

Figure 1: flow diagram of the proposed method.

Consider an image ‘X’ in which, the computational method is applied to adjust the intensity at dark region but globally. The same image is sent through the procedure,11 through which the brightness and contrast is adjusted. Consider the output image through computational method is XV and the output through is X L. A variation factor in intensity is calculated between the two images is given as follows:

The reconstructed image ’R’ is calculated as:

R=P* XV + (1-P) * XL (8)

The work is also extended to enhance the edges of the output image. All the channel of the reconstructed image is passed through the average filter. The edge is sharpened using the following equation as follows:

R1(Edge sharpened image)=R-(?*filtered image)/ß (9)

Where, ? and ß are the empirically defined parameter.

RESULTS AND DISCUSSION: In the section, some standard dataset has been taken to verify the applicability of the proposed method in the experimental analysis.

In order evaluate the visual quality of the image; we have applied this proposed method into some standard dataset. In Figure 2 first row shows the original input image. It can be shown that some image the quality is degraded locally and some images, the images contrast is affected locally. The Door and sunset images are affected globally due to low visibility. The image that contain teddy is affected locally due to shadow near by the object. It can be observed that the proposed method effectively enhanced the image at global and local level. Along with qualitative analysis, the quantitative parameter is also evaluated to examine the efficiency of the propose method. A quantitative parameter color root mean enhancement measure (CRME) is calculated to indicate the color image contrast quality. Higher the value of color root mean enhancement measure better would be the image quality.13-17

Table 1 CMRE of the input image.

Object

Proposed method

Teddy

0.3058

Wood

0.2631

Door

0.2122

Sunset

0.3118

peppers

0.2013

Figure 2: Restored image by different methods (a) Original Image (b) Histogram equalization method(c) Method (d) V-transform method (e) Proposed method. (f) Proposed method with sharp filter.

Table 2 : Comparison of average CMRE between other methods and propose one on the images available in datasets.

Method

Histogram equalization

Method18

Method10

Proposed method

Average Quality relative contrast measure

0.1412

0.1324

0.2197

0.2666

In table 1, it can be observed that the individual CMRE is achieved better through the proposed method. This suggests that the proposed method produces image which contains good contrast quality and better tonal retention. In table, it can be observed that the average CMRE achieved through the proposed method has achieved good contrast quality than others.

CONCLUSION: In this work, we have proposed the efficient image enhancement technique that enhances the contrast of image while keeping the brightness level optimum at local and global level. We utilized the computational photography method to achieve the optimum intensity level at the output. A weight is calculated and provides to computational photography image and locally improved contrast image to optimize the correct pixels at the reconstructed image. Experimental results show that the proposed method achieves the better contrast improvement ration than other existing state-of-the-art image enhancement methods.

REFERENCES:

1. Z. Liang, W. Liua, R. Yao (2015) Contrast Enhancement by Nonlinear Diffusion Filtering, IEEE Trans. Image Process., DOI 10.1109/ TIP.2015.2507405.

2. W. Fan, K. Wang, F. Cayre, and Z. Xiong (2015) Median Filtered Image Quality Enhancement and Anti-Forensics via Variational Deconvolution, IEEE Trans. on Inf. Fore. and Security, 10(5), 1076-1091.

3. M. Niknejad, H. Rabbani, and M. B. Zadeh (2015) Image Restoration Using Gaussian Mixture Models With Spatially Constrained Patch Clustering, IEEE Trans. on Image Process., 24(11), 3624-3636.

4. Kuang-Tsu Shih and Homer H. Chen (2016) Exploiting Perceptual Anchoring for Color Image Enhancement, IEEE Trans. on Multimedia, 18(2), 300-310.

5. Y. Niu, X. Wu, and G. Shi (2016) Image Enhancement by Entropy Maximization and Quantization Resolution Upconversion, IEEE Trans. On Image Process., 25(10), 4815-4828.

6. A. Jabeen, M. M. Riaz, N. Iltaf, and A. Ghafoor (2016) Image Contrast Enhancement Using Weighted Transformation Function, IEEE Sensors Journal, 16(20), 7534-7536.

7. Tariq Mahmood Khan, Donald G. Bailey, Mohammad A. U. Khan, and Yinan Kong (2017) Efficient Hardware Implementation For Fingerprint Image Enhancement Using Anisotropic Gaussian Filter, IEEE Trans. on Image Process., 26(5), 2116-2126.

8. S. Gu, W. Zuo, S. Guo, Y. Chen, C. Chen, and L. Zhang (2017) Learning Dynamic Guidance for Depth Image Enhancement,"in CVPR 2017.

9. S. W. Jung (2014) Image Contrast Enhancement Using Color and Depth Histograms, IEEE Signal Process. Lett., 21(4), 382-385.

10. Kim, Daeyeong, and Changick Kim. (2017) Contrast Enhancement Using Combined 1-D and 2-D Histogram-Based Techniques. IEEE Signal Processing Letters, 24(6), 804-808.

11. Carmona Salazar, O. D. and Calderon Gonzalez, J., (2015) Image enhancement with Matlab algorithms (Universitat Politècnica de Catalunya).

12. Tao, L., Seow, M. J., & Asari, V. K. (2006) February). Nonlinear image enhancement to improve face detection in complex lighting environment. In Electronic Imaging 2006 (606416-606416). International Society for Optics and Photonics.

13. K. Panetta, C. Gao, and S. Agaian (2013) No reference color image contrastand quality measures, IEEE Trans. Consum. Electron., 59(3), 643–651, Aug. 2013

14. S. Tiwari, K. Chauhan, and Y. Kurmi (2015) Shadow detection and compensation in aerial images using MATLAB, Int. Jour. of Comp. Appl., 119(20), 5-9.

15. Y. Kurmi and V. Chaurasia (2014) Performance of haze removal filter for hazy and noisy images,” Int. Jour. of Sci. Engg. and Tech., Apr. 2014, vol. 3 no. 4, pp. 437-439.

16. M. K. Patle, B. Chourasia, and Y. Kurmi (2016) High Dynamic Range Image Analysis through Various Tone Mapping Techniques, Int. Jour. of Comp. Appl.,153(11), 14-17.

17. A. Kumar, B. Chourasia, and Y. Kurmi (2016) Image defogging by multiscale depth fusion and hybrid scattering model, Int. Journal of Computer Appl., 155(11), 34-38.

18. T. Celik and T. Tjahjadi (2011) Contextual and variational contrast enhancement, IEEE Trans. Image Process., 20(12), 3431–3441.