Project Specification

     1. Project Overview

To separate or classify the cells by their given conditions automatically, one can extract the characteristics of cells in biomolecule fluids based on image processing analysis. These processes need to deal with a huge amount of image data in a high speed as fluid flowing, and must be done in very short time to control the separator. The purpose of this project is to implement a real-time system to compute them and get the cellular analysis by using Graphic Processing Unit(GPU)

2. Project Approach – Technical details

In cell biology, the image processing can be adapted the automatic analysis of cell characteristics and this requires the complex computation with high resolution of image data. Computational complexity makes analyzing system hardly implemented in real time, and the repetition in each pixel causes a heavy burden in utilization of hardware system.  Parallelization of image processing algorithm allows reduce computational time. GPU contains a large number of microprocessors and its architecture is very suitable for parallelization. Specialized hardware, such as FPGA or DSP, may work perfectly in the purpose of real time system. However, they needs specialty and expertize to be used as well as developed.  Therefore, GPU processing can be considered as an appropriate solution for the real-time image processing system

3. Project Objectives, Milestones, and Major Deliverables

> Project goals in short term

  • To optimize the algorithm for to get the most efficiency
  • To modify the process for parallelization
  • To implement the real time system using GPU platform

> Project goals in long term

  • To compare the efficiency of the implementation of GPU system to FPGA
  •  To characterize the property of the parallelizable module
  • To make a function for image processing analysis in parallel system

> Milestones and Major Deliverables


    4.     Constraints, Risk, and Feasibility

> Constraints

  •  The limitation of parallelization, because of data-dependency
  • The computational complexity needing time consumption

> Risk

  •  To work on only given test images
  •  To generate a code with redundancy

        5. Project Development





Desktop computer

w/ Graphic Processing Unit, Graphic Card by NVIDIA



CUDA toolkit



    6.     Project Schedule

  • Algorithm analysis based on MatLab code : To comprehend the basic algorithm of cellular image analysis based on MatLab code and literature survey
  • Conversion to C code and optimization of the algorithm : To convert MatLab code to C and modify some parts to be optimized for computational efficiency and parallelization
  • Parallelization in CUDA : To implement GPU code for real-time system in each module
  • Optimization in GPU : To optimize the result of parallelization to maximize the utilization rate of GPU hardware and minimize the redundant complex computations.