Dr Myat Su Hlaing Asia Research Center, Yangon University, Myanmar. Data programming model for an operation based parallel image processing system

Transcription

1 Name: Affiliation: Field of research: Specific Field of Study: Proposed Research Topic: Dr Myat Su Hlaing Asia Research Center, Yangon University, Myanmar Information Science and Technology Computer Science and related fields Data programming model for an operation based parallel image processing system Motivation and Need for the project A very large size digital image on a single machine can take a lot of time and the operation is computationally intensive. A possible solution is to reduce the processing time, to parallelize the processing on Central Processing Unit (CPUs). The aim is to take the advent of multicore CPUs and manycore Graphics Processing Unit (GPUs). The motivation for a parallel implementation of image algorithms comes from image and image sequence analysis needs posed by various application domains which are becoming increasingly more demanding in terms of the detail and variety of the expected analytic results, requiring the use of more sophisticated image and object models and of more complex algorithms, while the timing constraints are kept very stringent. The research is intended to divide large scale programs into several smaller programs for parallel execution by manycore GPUs with widely varying numbers of cores in order to reduce processing times. Background Parallel computing has attracted many of the researchers in recent years, who are trying to increase the performance of various applications and algorithms with the use of parallel computing techniques. Parallel computing is being used in a number of scientific and industrial applications from nuclear science to medical diagnosis. In computer science, it is being used for image processing, graphics rendering, data mining and various other applications. All these applications require large computation. Due to increase in computing power and storage of computers, demand for fast processing in increased. In general, parallel computing is the use of multiple computing resources simultaneously to solve a problem. It aims to solve a problem by dividing the problem into discrete sub-problems that can be solved concurrently. Image processing is an important part of the information work-out in the computer system. Image processing is the use of computer graphics algorithms to enhance the quality of digital images or to extract information about their content. Image filtering allows performing basic image editing tasks such as image smoothing, sharpening, blurring, edge detection, mean removal and embossing. Some parallel image processing that has been actively used by research communities. Algorithms for primary image processing are noise reduction and an overall image enhancement. Algorithms for an intermediate image processing are used for segmentation of image, defining of skeleton, edge detection, etc. The filter mean belongs to the group of algorithms for primary image processing. It is also known as neighborhood averaging. The image processing tasks are identified with operations of the same kind held over large data massive. The tremendous amount of data required for image processing and computer vision applications presents a significant problem for conventional microprocessors. The goal of this research is to reduce these large-scale efforts focused on the use of such a high performance

2 computing system as highly parallel multithreaded environment of GPU using Compute Unified Device Architecture (CUDA). Objectives of Research The purpose of the research is to present an application of a data-parallel programming model for an operation based parallel image processing system. The user of a computationally demanding application may benefit from the computational power distributed on the advent of multicore CPUs and manycore GPUs. The main objectives of my research are mentioned: (1) To present a multithreading method for computing the image registration computing (2) To parallelize the image registration using multiprocessing method (3) To reduce the processing time and (4) To speed up the computations. Scope, Design and Methodology Scope of Research Image is the two dimensional distributions of tiny image points called as pixels. It can be considered as a function of two real variables, for example, f(x,y) with f as the amplitude (e.g., brightness) of the image at position (x,y). Image Processing is the process of enhancing and manipulation with an image in order to extraction of meaningful information. Image processing has become a useful research area that goes from professional photography to several different fields such as Astronomy, Computerized photography (e.g., photoshop), Space image processing (e.g., Hubble space telescope images, interplanetary probe images), Medical/Biological image processing (e.g., interpretation of Xray images, blood/cellular microscope images, CT Scan, PET Scan), Automatic character recognition (zip code, license plate recognition), Finger print/face/iris recognition, Remote sensing: aerial and satellite image interpretations, Reconnaissance, Industrial applications (e.g., product inspection/sorting). Image smoothing filters are used for blurring and for nose reduction. Blurring is used in preprocessing steps, such as removal of small details from an image prior to object extraction, and bridging of small gaps in lines or curves. Noise reduction can be accomplished by blurring with a linear filter and also by non-linear filtering, order-statistic filters. This has the effect of eliminating pixel values which are unrepresentative of their surroundings. Main principle of parallel computing is to divide a task in such a way that the task executes in minimum time with maximum efficiency. To implement parallel computing there can be several kind of parallel machine like a cluster of computers which is having multiple PCs combined together with an elevated speed network using Message Passing Interface (MPI); a shared memory multiprocessor by connecting multiple processors to a single memory system Open Multi-Processing (OpenMP), hybrid of OpenMP and MPI. For implementing this in image processing, several research and contributions have been done parallel processing with several tool like GPU using CUDA, Java, Hadoop and OpenCV and MATLAB (2014) and etc. However, it is very important to find most suitable technique of parallel computing for a particular application of image processing. Design of Research Parallel implementation of image algorithms comes from image and image sequence analysis needs posed by various application domains which are becoming increasingly more

3 demanding in terms of the detail and variety of the expected analytic results, requiring the use of more sophisticated image and object models and of more complex algorithms, while the timing constraints are kept very stringent. The goal of this research is to reduce these large-scale efforts focused on the use of such a high performance computing system for scientific applications. The purpose of the research is to present an application of parallel computing for an operation based parallel spatial image filters system. The main objectives of this system are to present a multithreading method for computing the image registration computing, to parallelize the image registration using multiprocessing method, to reduce the processing time and to implement spatial image filters using multithreading and multiprocessing methods of CPUs and GPUs. Methodology of Research Most numerical approaches for image registration are based on the computing over the pixel matrices. As the image size increases, more time consuming is needed. Image parallel registration is employed for the data-parallel processing approach to reduce the execution time. This research is focused on spatial filtering operations for image enhancement. Smoothing Spatial Filter Methods is discussed by using smoothing linear filters and order-statistics filters. The linear image filtering methods are: 1. Arithmetic Mean Filter The idea of Arithmetic mean filtering is simply to replace each pixel value in an image with the mean value of its neighbors, including itself. This has the effect of eliminating pixel values which are unrepresentative of their surroundings. 2. Harmonic Mean Filter When harmonic mean filter is applied to an image, the color value of each pixel is replaced with the harmonic mean of color values of the pixels in a surrounding region. 3. Contraharmonic Mean Filter Contraharmonic Mean Filter is well suited for reduction or virtually eliminating the effects of salt-and-pepper noise. For positive values of q the filter eliminates pepper noise. For negative values of Q it eliminates salt noise. Image order-static filters are nonlinear spatial filters whose response is based on ordering, ranking the pixels contained in the image area encompassed by the filter, and then replacing the value of central pixel with value determined by the ranking result. 1. Median Filter In order to perform median filtering at a point in an image, the median is calculated by first sorting all the pixel values from the surrounding neighborhood into numerical order and then replacing the pixel being considered with the middle pixel value. 2. Midpoint Filter The Midpoint filter blurs the image by replacing each pixel with the average of the highest pixel and the lowest pixel within the specified window size. 3. Alpha-trimmed mean filter Alpha Ttrimmed mean filter is windowed filter of nonlinear class by its nature is hybrid of the mean and median filters. Both sequential implementation methods and parallel implementation methods are described for spatial image filtering methods. Two parallel implementation methods are discussed to perform the Smoothing Spatial Filter Methods. These parallel approached are by

4 processing with the shared memory implementation using OpenMP, highly parallel multithreaded environment of GPU using CUDA. 1. Shared memory implementation using OpenMP The Open Multi-Processing (OpenMP) is an application programming interface (API) that supports multi-platform shared memory multiprocessing programming in C/C++ and Fortran on many architectures, including Unix and Microsoft Windows platforms. It consists of a set of compiler directives, library routines, and environment variables that influence run-time behavior. OpenMP is now used by many software developers; it offers significant advantages over both hand-threading and MPI. Using OpenMP offers a comprehensive introduction to parallel programming concepts. 2. Multithreaded environment of GPU using CUDA CUDA is a parallel computing platform and programming model invented by NVIDIA. It enables dramatic increases in computing performance by harnessing the power of GPU. CUDA programs are compiled by nvcc compiler provided by NVIDIA. It first separates the host and device code from a CUDA program. The host code is then further compiled by a standard C compiler like gcc. While nvcc further compiles the device code as well to make it able to execute in parallel on GPU. There are four memory types in CUDA, Global Memory, Shared Memory, Constant Memory and Registers. Each has different access latency and bandwidth. To achieve best performance gain, each memory type should be used efficiently. Schedule of Research The following table shows the schedule of my Research. Month Description 1 st Month Image Linear Filterings with Shared memory implementation using OpenMP 2 nd Month Image order-static filtering with Shared memory implementation using OpenMP 3 rd Month Image Filtering with of GPU using Global Memory Types in CUDA 4 th Month Image Filtering with of GPU using Shared Memory Types in CUDA 5 th Month Image Filtering with of GPU using Constant Memory Types in CUDA 6 th Month Image Filtering with of GPU using Registers in CUDA Significance This research will focus on Image Filtering Methods and Parallel Computing Development. The main problem of large scale computation of image processing is that it is generally time consuming process; Parallel Computing provides an efficient and convenient way to address this issue. The aim is to implement parallel computing in image spatial filtering applications. The non-linear image filtering and image order-statics filters are intended to develop by parallelization methods with a shared memory implementation (OPenMP) and GPU using CUDA. This research will reduce the execution time of the program and make a comparison of GPU using CUDA version and a sequential CPU version. Dissemination Plan This proposed project represents an attempt to understand parallelized methods of OPenMP and GPU using CUDA. After finishing the research project, I will be able to develop many of the ideas that will extend from the research experience in Korea. I will implement

5 some ideas in our University and our Ministry to improve research of Parallel Computing and Image Processing in Myanmar. Moreover I will transfer the knowledge that I will gain in Korea University to my student and share the knowledge to my colleague. Finally, I will try to contribute extensive research experience in many areas and can easily modify my research program to suit the needs of Department of Computer Studies, University of Yangon as well as my country. References (1) Cavarra, A., Caramagno, D., Parallel Computing Solutions for Linear Combination of Filters, University of Catania, Viale A. Doria 6, Catania, Italy, 2016 (2) Grama,A. and Gupta,A. and Karypis,G. and Kumar,V., Introduction to Parallel Computing, 2 nd Edition, Pearson Education, The Benjamin/Cummings, ISBN: , 2003 (3) Nickolls, J., GPU Parallel ComputingArchitecture and CUDA Programming Model, Hot chips 2007: NVIDIA GPU parallel computing architecture, NVIDIA Corporation (4) Parhami,B., Introduction to Parallel Processing Algorithms and Architectures, Plenum, New York, ISBN , 2002 (5) Saxena, S., Sharma, S. and Sharma, N., Parallel Image Processing Techniques, Benefits and Limitations, Research Journal of Applied Sciences, Engineering and Technology, January 2016 Journal Publication Publications List of Dr Myat Su Hlaing (1) Than Zaw Nyunt, Myat Su Hlaing, "The Analysis of Direct DCT Image Indexing for JPEG Compression", Journal of The Myanmar Academy of Arts and Science, Yangon, Myanmar, 2007, Vol.V, No. 3 (2) Myat Su Hlaing, Than Zaw Nyunt, "The Analysis of Image Enhancement by Filtering", Journal of The Myanmar Academy of Arts and Science, Yangon, Myanmar, 2008, Vol.VI, No.3 (3) Soe Mya Mya Aye, Myat Su Hlaing, Htway Htway Khaing, "Transformation of one Font to Another Myanmar Font", Journal of The Myanmar Academy of Arts and Science, Yangon, Myanmar, 2008, Vol.VI, No.3 (4) Myat Su Hlaing, "Parallel Computing Based On Cluster Computer By Using OpenMP And MPI", Universities Research Journal, Yangon, Myanmar, 2009, Vol.2, No.3 (5) Myat Su Hlaing, "Data Parallelism in One Dimensional Fast Fourier Transform", Universities Research Journal, Yangon, Myanmar, 2010, Vol.2, No.1 (6) Sann Htoo, Myat Su Hlaing, Than Than Wai, Ye Chan, "Parallel processing in Cryptography", University of Yangon Research Journal, Yangon, Myanmar, 2011, Vol.2, No.1

6 Conference Presentation (1) Myat Su Hlaing, Than Zaw Nyunt, "The Analysis of Image Enhancement by Filtering", Paper Reading Section of The Myanmar Academy of Arts and Science, Yangon University, Ministry of Higher Education, 2008 (2) Myat Su Hlaing, " Parallel Computing Development for Image Order-Statistics Filters", Research competition, Yangon University, Ministry of Higher Education(Lower Myanmar), 2013 (3) Myat Su Hlaing, Soe Mya Mya Aye, Pho Kaung, " Parallel Computing Development for Image Order-Statistics Filters", Proceeding of The Sixth International Conference on Science and Mathematics Education in Developing Countries, Mandalay, Myanmar, November 2013 (4) Myat Su Hlaing, "Parallel Image Order-Statistics Filters Using Shared Memory and Distributed Memory", Proceeding of The 3 rd International Conference on Computer Applications and Information Processing Technology (CAIPT 2015), Yangon, Myanmar, June 2015 (5) Myat Su Hlaing, "Comparison of Computing Time among Parallel Image Spatial Linear Filters", Proceeding of The Eighth International Conference on Science and Mathematics Education in Developing Countries, Yangon, Myanmar, November 2015 (6) Myat Su Hlaing, "Parallel Computing Development for Image Smoothing Linear Filters", Research competition, Yangon University, Ministry of Higher Education(Lower Myanmar), 2015 Proceeding Papers (1) Myat Su Hlaing, Soe Mya Mya Aye, Pho Kaung, " Parallel Computing Development for Image Order-Statistics Filters", Proceeding of The Sixth International Conference on Science and Mathematics Education in Developing Countries, Mandalay, Myanmar, November 2013, pp (2) Myat Su Hlaing, "Parallel Image Order-Statistics Filters Using Shared Memory and Distributed Memory", Proceeding of The 3 rd International Conference on Computer Applications and Information Processing Technology (CAIPT 2015), Yangon, Myanmar, June 2015, pp (3) Myat Su Hlaing, "Comparison of Computing Time among Parallel Image Spatial Linear Filters", Proceeding of The Eighth International Conference on Science and Mathematics Education in Developing Countries, Yangon, Myanmar, November 2015 Dr Myat Su Hlaing Lecturer Department of Computer Studies Yangon Myanmar