Computer Vision System for Image Quality Assessment - Research Proposal Example

Running Head: Computer vision SYSTEM for image quality assessment Computer vision system for image quality assessment Introduction Image quality assessment is the process of measuring the perceived degradation of an image with or without a baseline image. The process of image quality assessment where a baseline image is used is called as the Full reference (FR) method. The baseline image is assumed to have a perfect quality. For example, when an original image is compared to the JPEG (Joint Photographic Experts Group) compression of it, then it is using the original image as the reference. The other process of image quality assessment is called as the NR or No-reference method. As obvious from the nomenclature, the NR process does not use a reference image for assessing the quality of an image. Image quality assessment is a significant issue as the modern imaging systems has a tendency to introduce artefacts or distortion in the signal (Wang, & Bovik, 2006). The entire process of capture and development of an image is a result of two steps. The first step is the capturing of image by a photographic process. The ideal photographic process is standardised with reference of a pinhole camera. The pinhole camera is a hardware apparatus, which has light reflected from the intended subject passing through the aperture and falling on the plane of image. In practical applications, the camera approximates the ideal pinhole camera image capture process. Ideally, the developed image should exactly capture the amount of light falling on the each part of the image plane of the camera, whether the image is developed chemically (traditional) or digitally (Modern). The second step consists of the processing of the amount of light falling on the image plane. Ideally, the second step should perfectly measure the amount of light falling on different parts of the image plane and hence reproducing the photograph exactly as per the depicted scene, whether on a paper or in the soft copy. In practice, some amount of distortion takes place in each step mentioned in the previous paragraph. The practical cameras are an approximation of the ideal pinhole camera. Thus, the amount of light falling on the image plane is not exactly as reflected by the depicted scene. Similarly, the processing of the image by the measurement of the amount of light in the different areas of image plane is not perfect. The image storing & transmission process may also distort the quality of image. For example, a digital image, which has been compressed, transmitted and decompressed, would not have same quality as the original. Thus, the measurement of the image quality would be the distortion in the final image with respect to the original image (Chandler, & Larson, 2010). In order to conduct an assessment of the image quality, the human eye itself acts as a camera. Extensive research has been conducted to model the process of human vision. Various parameters like colour vision, limitation of human eye, perception of the scene etc, have been studied. The research on the function of the human eye has been applied as visual models on the video quality and the associated metrics. The aim of these applications has been to create and optimize the compression/encoding/decoding the algorithms of MPEG-4 & MPEG-2. Metrics have been developed to examine the probability of detecting artefacts in the static and the moving images. These are also known as threshold metrics (Rao, & Rao, 2006). Thus, there are various image assessment techniques to rate the quality of images. The term image quality assessment includes various methods, which are used by quality personnel for categorising or ranking the printed or digital images. This project is an exercise to develop a system to assist the QA personnel in ranking the images. Thus, the goal of the project is to construct a computer vision system that would facilitate image quality assessment tasks performed with developed prints or photographs on a table. The project will involve constructing a computer vision system, in which a video camera would be positioned above the table and pointing to the surface of the table. This video camera would focus on the images, which would be laid out on the table for the purpose of assessment. Each print that needs to be assessed will have a marker in one of the corners that will be detected by the camera. The system will register the position of the pictures on the table based on the position of the markers. The assessment of the images would go on until stopped by a signal, which is built-in the system. This signal would be initiated by the process of ‘virtual’ button. The virtual button is another marker that needs to be covered with hand for the system to ‘understand’ that the current set of photographs is finished. Subsequent to this action, the user can begin the assessment of the next set of photographs. Motivation for the project. This project is designed with an objective to provide an automated image assessment process to the QA professionals in this area. Usually, visual examination of the images is conducted to determine its quality. The person doing the visual check may also be provided with a checklist to rank the images. The checklist could be based on an existing process. For example, an image that is stored in JPEG file with low quality settings will look worse than the original image and hence, will have a lower quality score (Bovik, 2009). One of the methods of estimating quality score of images involves ordering or ranking the printed images on a table. The images are laid out in the order with the highest quality image on the left, to the image of the poorest quality on the right. The advantage of this method is that it is intuitive process and hence, quite efficient. The quality assessor can follow this protocol to make a judgement on the image quality. The next step is to register the results obtained from the assessment. If the assessor is not careful to record his assessment into the printed image, or if he does not record it onto the log sheet, there is a possibility of the loss of data. Another important factor is that the assessment of an image is against the set of images, thus the entire stack needs to be held together to retrieve the correct results of the image assessment. This issue has high probability of occurrence in case, the assessor and record keepers are not same. If there are training gaps in terms of understanding the rank codes on the images, the record keeper may, unintentionally, mix up the results. Thus, the disadvantage of this process is that it is time consuming and monotonous, making it prone to errors & fatigue. On the other hand, if the assessor himself is logging the results then it’s wastage of his precious time. He should be ideally spending time in assessing the images, for which he has an expertise. The logging of results can be done by anyone, who has some experience in data entry and has been trained into understanding the image rank code standards. Thus, as highlighted above, the image assessment is a subjective process and needs a trained person for its efficient execution. The result registration process is monotonous and needs automation to improve the quality of recording results and reduce manual errors. Aims and objectives. The project aims to automate the process of the image assessment and registration of the results of assessment. This aim would be achieved by using the new computer vision tool for assessing images in a stack. This tool would operate as per the process given in the earlier section. As the image assessment is a subjective process, it is important to do a primary study for arriving at the success of new process. The study needs to be conducted with certain success parameters in mind so that results can be recorded without any bias. The parameters of success for this project are: Reduction in the time of registering the image assessment results, making the entire process easier & easily transportable across resources and a ‘virtual’ button stopping the process of assessment without fail. As the existing baseline process parameters are not officially documented, a primary study would be conducted to report the metrics of the existing and new automated process. The primary group of participants would include 10-15 individuals, who would be asked to assess the 5-10 images using the new tool. The results of this assessment would be compared against the assessment done without the use of the new tool. Organization of this dissertation This dissertation is divided in four chapters, with each chapter having multiple sections. The high level logical distribution begins with providing the background of the case, followed by literature review & primary case study. Finally, the discussions on the results are conducted with appropriate derived conclusions. The dissertation closes with the future scope of research in this domain of automating the process of image assessment and registration. The Chapter 1 Introduction explains the motivation behind this research, the aim & objectives along with laying down the critical success parameters. Each success parameter is provided with a threshold value, which must be achieved to judge the project as success. Chapter 2 Literature Review discusses the work done by other researchers and authors in the field of Image assessment and particularly on automating the process of assessing and recording the image ranking. This chapter would also delve deeper on the area of subjective quality assessment, various user interfaces using augmented reality and the process behind the tracking of markers. Chapter 3 Computer vision system for quality assessment is devoted to the intricacies of the computer vision hardware and software design. This chapter would discuss the system requirements, testing needs, the case study details and finally, it would tabulate the results of the experiment. The last and the fourth chapter of this dissertation, discussion and conclusion, summarises the result obtained from the study and develops conclusions by analysing the success parameters. The conclusions would provide the final assessment of the success and failure of this new tool. This chapter would also provide the scope of further research in this domain. The further research can be done to improve the issues with the current tool or it can be in the area of enhancing the tool further to handle more requirements. References Bovik, Alan. (2009). The Essential guide to image processing. 2nd Ed, Academic Pr. Chandler, Damon M, & Larson, Eric C. (2010). Most Apparent distortion: full-reference image quality assessment and the role of strategy. J. Electron. Imaging, 19(011006 (2010)) Rao, Kamisetty, & Rao, K.R. (2006). Digital video image quality and perceptual coding. CRC Press. Open Source Computer Vision, Initials. (2010, August). OpenCV wiki. Retrieved from http://opencv.willowgarage.com/wiki/ Wang, Zhou, & Bovik, Alan. (2006). Modern image quality assessment. Morgan & Claypool. Read More