Quality Measure of Multicamera for Geometric Distortion Mahesh G. Chinchole 1, Prof. Sanjeev.N.Jain 2 M.E. II nd Year student 1, Professor 2, Department of Electronics Engineering, SSVPSBSD College of Engineering, Dhule North Maharashtra University, Jalgaon, Maharashtra, India Abstract To create the multiple events into the single image is a simple way to look all the events in a single look. For this multicamera images have to combine into single image also known as multi-view image. When we combine the multiple images into single image, due to misalignment and different camera orientation as well as arrangement the geometric distortions comes into picture. The proportional distortion variation between two separate camera images is the main aspect while measuring the required quality of the final image. So the quality measure or quality analysis is the most important (step) or factor in multicamera images. There are several objective & subjective methods have been proposed for single camera images but no such comparable efforts has been taken on multicamera image quality measure. This paper details the methods and results of implementing MIQA multicamera image quality analysis. Here we show the methods like PSNR, MSSIM, and VIF for the measurement of quality of multicamera image and then compare their results with MIQM. The experimental analysis shows the effectiveness of the every method in comparison with others. Here we consider the 1 value for original or reference image and 0 values for complete distorted image. The range of MIQM is ranging from 1 to 0 Keywords PSNR, MSSIM, VIF, MIQM, quality assessment, Full reference, reduced reference, No reference. I. INTRODUCTION The image quality can be measured interms of either subjectively or objectively [1] [2] [3]. The objective image quality measure means that the image quality is measured either by some methods, techniques or by algorithm. In contrast, in subjective image quality measure the peoples are asked for their opinion on the image quality because in most cases human eyes are the decisive receivers. The mean opinion score (MOS) [4], provides a mathematical analysis of the apparent quality of an image and is obtained from a number of human being observer. This method has been used from many years; but, the MOS is monotonous, quite expensive in terms of time and human resources. Additionally, the subjective quality measures results depend on numerous external factors for example the observer s surroundings, interest motivation, etc [5]. So it shows that the manipulation is more possible in case of subjective measure. In objective image quality measure, the distorted image is compare with the reference image. In a simple way the, the analysis is done by subtracting the reference image from the distorted image which leads to the mean square error (MSE) or Peak signal to noise ratio (PSNR) [1][2][3]. The aim of objective image quality measure is to predict the perceived image quality by human vision model which can predict the perceived image quality without human intervention. The objective measure should provide the mathematical value to the persons having dissatisfaction when they observe the reproduced image instead of reference [5]. As per the availability perfect reference image that is supposed to have ideal image quality, the image quality measures can be classified into full reference (FR), reduced reference (RR), and no-reference (NR) methods. In FR measure, the full access to the reference image is available, while in NR no access to the reference image. The RR, the partial access is available to the reference image. The access means amount of information or features extracted from the reference image is available for the assessment of the quality of the distorted image [6]. To measure the quality of the distorted image, FR methods [7] provide the most accurate results in contrast with RR & NR. The conventional Full reference image quality assessment methods calculates peak signal-to-ratio (PSNR) and Mean Square Error (MSE) i.e. the pixel-wise distances between a distorted image and reference image [8]. The digital images are subject to a variety of distortions that may result in degradation of visual quality throughout acquisition, processing, compression, storage, transmission and reproduction. So, it is essential for numerous applications to be able to measure the image quality degradation that occurs in a system [5]. The objective of this paper is to introduce the objective quality measure for geometrically distorted multicamera images, which has the better connection with human observation of the distortions. The basic idea of our approach is that the human vision is extremely perceptive to changes in structures in images, thus structural distortion should be a good estimation of the supposed image distortion. In full reference approach we have the reference image. Then for multicamera image quality measure we create the geometrically distorted image. After that we compare both the images by various quality measure techniques. In the second comparison we use the Barbara standard images which are local geometric distortions applied from [5]. Then we performed the analysis which shows the success of our approach in contrast to others. II. GEOMETRIC DISTORTION IN MULTI CAMERA IMAGE SYSTEM The geometric distortion can be defined as the shifting of pixels or overlapping of pixels on each other in the image. For a multicamera image the particular scene is to captured by number of cameras from different positions or by different angles. The example of such system is shown in figure.1. ISSN: 2231-5381 http://www.ijettjournal.org Page 155
a) Parallel. b) Convergent c) Divergent Fig. 1 Three possible camera configurations, i.e., parallel view, convergent and divergent view. The fig. 1 shows the three different cameras configurations which are placed to capture the event. As due to different camera positions with different angles in short with different orientations with camera calibration parameters [9] the geometric distortions will be created in creating the multicamera images. The Geometric distortion can be divided in two ways as linear geometric distortion and angular geometric distortion. The linear geometric distortion occurs during the rotation, translation, in motion. In this position of the pixels gets shifted or overlapping of pixels is happened. The angular geometric distortion occurs during the mapping like 3-D plane to 2-D plane. The examples of linear and angular geometric distortion are given below [10] [11]. (a) Original (c) Linear Distortion. (b) Original (d) Angular Distortion (rotation). Fig.2. Example of geometric distortion in multicamera images. Fig.2 shows the examples that demonstrate the types of geometric distortion with original image. The image in fig.2.(d) is cause to undergo the angular (rotational) distortion. The columns look closer compare to original image shown in fig. (b). in fig. (d) rotated by angle 3 degree clockwise. In multicamera system such distortion can occur when there is mapping of certain camera plane to reference camera plane. In single-view images the geometric distortions have been considered in [12]. The authors proposed the complex wavelet domain image comparison which is unresponsive to spatial translations. The proposed model considered that the single view image is insignificant due to perceptual distortions caused by spatial scaling, rotation, and translation. Though, this consideration is not true for multiview images, where discontinuities, misalignments, blur, and double imaging can effect in catastrophic distortions. Thus, a precise multicamera image quality assessment must report for geometric distortions [12] [9]. I. QUALITY ASSESSMENT OF MULTICAMERA IMAGES The image analysis is concern with the extraction of measurement, data or information from an image by automatic or semiautomatic methods. The image analysis is distinguished from other types of image processing such as coding, restoration, and enhancement. In image analysis, the ultimate output is usually numerical output rather than picture or image. [11][13]. The techniques used for extracting information from an image are known as image analysis techniques or image quality measurement techniques. An image composed of edges and shades of gray. Edge is corresponding to fast change in gray level and thus corresponds to high frequency information. Shade is corresponds to low frequency information. Separation (filtering) of high frequency information means edge detection. An edge or boundary is the external information of image. The internal features in an image can be found using segmentation and texture. These features depend on the reflectivity property. Segmentation of an image means separating certain features in the image. While treating other part as a backdrop if the image consists of a number of features of interest then we can segment them separately. Texture of an image is quantitatively described by its roughness. The roughness index is related to the spatial repetition period of the local structure. It is necessary to segment the image based upon uniform texture before its measurement. feature is a distinguishing characteristic of an image. Spectral and spatial domain is the main methods used for feature separation Motion of an object studied from study of multiple images, separated by varying periods of time[11] [14]. There are several applications of multicamera system and every application has precise meaning for post processing and presentation [15]. In these a single camera is generally chosen as reference camera for estimating plane or geometry [16]. Here we are presenting the full reference system and estimating the image quality for multicamera system. To simulate distortions in multicamera images, a single digital camera was used to capture high-resolution images. Each image was then divided into various sub-images. Then these images are then combining with some % of overlap. The overlap areas where different with each image; yet, they were all in the range of 5%, 20% & 40% of the original image. After that the quality measurement of the distorted image is done by different quality methods like PSNR, VIF, MSSIM and finally with our approach MIQM. ISSN: 2231-5381 http://www.ijettjournal.org Page 156
I. PROPOSED SYSTEM ARCHITECTURE As we know the multicamera images are suffers from mainly two types of distortion Geometric distortion and Photometric Distortion. So we obtained the results by considering the individual distortions. Here we simulate the single camera images for the geometric distortion in case of Multiview. B) LINEAR GEOMETRIC DISTORTION - To simulate the geometric linear distortion in multicamera system, the geometric distortions were applied to the images shown in fig. 3. In this we Split image into two parts (Left and Right). Here we created the 3 different images having different amount of overlap. The amount of overlap should be a design parameter. In the first image we overlap the two parts by 5% and it is named as low overlap image and then calculated the PSNR, VIF, MSSIM and finally the MIQM. In the second image we overlap the two parts by 20% and it is named as medium overlap image and then calculated the PSNR, VIF, MSSIM and finally the MIQM. In the third image we overlap the two parts by 40% and it is named as high overlap image and then calculated the PSNR, VIF, MSSIM and finally the MIQM. Fig.3 shows the three examples of geometric distortion in multiview images. Fig. 3 (a) is an original image. (a) Original (b) Low (5 % overlap) PSNR = 1, VIF = 1, PSNR = 0.3422, VIF = 0.9857, MSSIM = 1, MIQM= 1 MSSIM = 0.8719, MIQM= 0.6573 II. EXPERIMENTAL RESULT The objective of MIQM is to obtain an innovative quality measurement for multi-camera images. As the quality of images are affected by many factors like number of cameras, camera configuration, calibration process, quality evaluation for such an image should take all these into consideration. To design an objective metric for multi-camera images, the visual distortion is identified into two types, photometric distortion and geometric distortion, which can be translated into luminance, contrast, spatial motion and edge-based structure components. The main thought of the paper is to first measure each component by proposing different index values and then to combine those indexes into one quality, MIQM, to capture the perceptual quality of multiview images. MIQM shows the measurement that is full -reference and designed to evaluate multi-view images, where the reference is regarded as the set of images taken by identical cameras. Different types of distortions can be calculated in MIQM according to the following method describe below. A) MIQM RESULTS: The results of MIQM are obtained by following steps. (c) Medium (20 % overlap) (d) High (40 % overlap) PSNR = 0.3196, VIF = 0.9501, PSNR = 0.3037, VIF = 0.9083, MSSIM = 0.6927, MIQM= 0.4059 MSSIM = 0.4893, MIQM= 0.1970 Fig.3. Linear Geometric distortion. (a) Original (b) Linearly overlaps by 5% (c) Linearly overlaps by 20% (d) Linearly overlaps by 40% s Quality Methods Original 5 % 20 % 40 % PSNR 1 0.3422 0.3196 0.3037 VIF 1 0.9857 0.9501 0.9083 MSSIM 1 0.8719 0.6927 0.4893 MIQM 1 0.6573 0.4059 0.1970 Table.1. Multicamera Quality Measure Methods Results for the geometrically distorted Multicamera s ISSN: 2231-5381 http://www.ijettjournal.org Page 157
(a) Original (a) Staircase Plot (b) Medium Distorted (b) Scatter Plot (c) High Distorted Fig.5 Geometrically distorted standard Barbara s. (a) Original (b) Medium Distorted, (c) High Distorted. (c) Bar Graph Fig.4. Quality Measure results in the form of (a) Staircase Plot, (b) Scatter Plot, (c) Bar Graph Fig. 4 shows the quality measurement for multicamera images with the methods like PSNR, VIF, MSSIM, MIQM interms of graph & plots like Bar graph shown in figure (c), Staircase plot and scatter plot shown in figure (a) & (b) respectively. Table 1 shows the numerical results of quality measure for the simulated multicamera image. If we observe the graph as well as table for comparative analysis of quality measure methods for multicamera images, it is observed that the MIQM shows the better result and sensitivity compare to other quality parameters. At the start of analysis when the overlap is less, the PSNR showing the better result compare to the MIQM but as we increase the amount of overlap the sensitivity of PSNR goes on decreasing at that time the MIQM shows the better result. We also tested our approach to the Barbara images available in [5] which are shown in figure 5, here also we obtained the better result in contrast to others. s Quality Methods Original Medium Distorted High Distorted PSNR 1 0.3043 0.3019 VIF 1 0.7820 0.8100 MSSIM 1 0.4599 0.4653 MIQM 1 0.1890 0.1754 Table.2. Comparative of Quality Measure Methods for geometrically distorted standard Barbara s. Table 2 shows the comparative analysis for the standard Barbara images with respect to different quality measure techniques. Here we measure the quality of medium distorted image & high distorted image with respect to original image. For original image all the values are 1 but as we increase the distortion the values gets decreases. Figure 6 shows the bar graph for the image quality measure of standard Barbara images. ISSN: 2231-5381 http://www.ijettjournal.org Page 158
Fig.6. Quality Measure results in the form of Bar graph for III. CONCLUSION AND FUTURE WORK In this paper, we proposed the full reference objective quality measure for assessment of the perceptual quality of multicamera image for geometric distortion. In the past only few works has been found in indicating the problem of geometric distortion for multicamera image. The proposed measure is based on the Luminance and contrast index, spatial motion index and edge based structural index. The Experimental results show that how it is giving the better results in contrast to PSNR, VIF and MSSIM. In the initial analysis the PSNR is giving the better result but as we are increasing the distortion the sensitivity of the PSNR decreases sharply but in that case our approach is indicating the better result as shown in the various graphs. In future we can use the databases like the images which are the combination of both photometric and geometric distortion and the images having the geometric distortion interms of pixel shifting. Quality Measure IEEE Transactions On Processing, Vol. 21, No. 9, pp 3902-3913, September 2012 [10] M. Solh and g. Alregib, characterization of image distortions in multi-camera systems,in proc. 2nd int. Conf. Immersive telecommunication., may 2009, pp. 1 6. [11] Mahesh G. Chinchole, Sanjeev N. Jain, A Review On Multicamera Quality Analysis International Journal of Engineering Research and General Science (IJERGS) Volume 3, Issue 3, May-June, 2015, ISSN 2091-2730 [12] Z. Wang and e. Simoncelli, translational insensitive image similarity in complex wavelet domain, in proc. Icassp, mar. 2005, pp. 573 576. [13] William k. Pratt, digital image processing, third edition, a wileyinterscience pubication, john wiley & sons, inc. New york 2006. [14] Madhuri a. Joshi, digital image processing, an algorithmic approach, eastern economy edition, phi learning private limited, new delhi 110001, 2010 [15] Applications and Requirements for 3DAV, ISO Standard N5877, Jul.2003. [16] C. Tang, c. C. Y. Yu, and c. Tsai, visual sensitivity guided bit allocation for video coding, IEEE trans. Multimedia, vol. 8, no. 1, pp. 11 18, feb. 2006. REFERENCES [1] Arend Eden- No-Reference Quality Analysis for Compressed Video Sequences IEEE Transaction on Broadcasting, Vol. 54, No.3, pp. 691-697, September 2008. [2] Methodology for the Subjective Assessment of the Quality of Television Pictures, Recommendation ITU-R Rec. BT. 500-11. [3] Z. Wang and a. C. Bovik, modern image quality assessment. New york: morgan & claypool, 2006. [4] Subjective Video Quality Assessment Methods for Multimedia Applications Recommendation P.910. Geneva, Switzerland, 1996, International Telecommunication Union. [5] Angela D Angelo, Li Zhaoping, and Mauro Barni, A Full- Reference Quality Metric for Geometrically Distorted s IEEE Transactions On Processing, Vol. 19, No. 4, pp. 867 880, April 2010 [6] Zhou Wang, Applications of Objective Quality Assessment Methods IEEE Signal Processing Magazine, pp 137-142, November 2011. [7] H. R. Sheikh, m. F. Sabir, and a. C. Bovik, a statistical evaluation of recent full reference image quality assessment algorithms, IEEE trans. process., vol. 15, no. 11, pp. 3440 3451, Nov. 2006. [8] Xinbo Gao, Wen Lu, Dacheng Tao, Xuelong Li, Quality Assessment Based on Multiscale Geometric Analysis, IEEE Transactions on Processing, Vol. 18, No. 7, pp. 1409-1423, July 2009 [9] Mashhour Solh, Ghassan alregib, MIQM: A Multicamera ISSN: 2231-5381 http://www.ijettjournal.org Page 159