200 likes | 354 Views
A Cell Image Segmentation Algorithm By Simulating Particle Movement. Project report of Computer Vision Xijiang Miao. Outline. Introduction Related works The algorithm Potential problems. Cells under microscope. mitosis. Gap. synthesis. apoptosis. Mission: Tell apart each cells.
E N D
A Cell Image Segmentation Algorithm By Simulating Particle Movement Project report of Computer Vision Xijiang Miao
Outline • Introduction • Related works • The algorithm • Potential problems
Cells under microscope mitosis Gap synthesis apoptosis
Mission: Tell apart each cells • Knowing the number of cell is helpful • Extract RNA, … • Currently, the number of cells is manually counted. • Classifying cells in different phase is valuable. • Check the effect a treatment. • Integrate into cell sorting machine.
A recent published vote based algorithm Yang, Q. et al, Perceptual Organization of Radial Symmetries, Proceedings of (CVPR’04)
Watershed algorithm Fig. 2. Building dams at the places where the water coming from two different minima would merge. Vincent, L. and Soille, P. Watersheds in Digital Spaces: An Efficient Algorithm Based on Immersion Simulations.IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 13, NO. 6, JUNE 1991
Watershed in ImageJ ImageJ: http://rsb.info.nih.gov/ij/ Watershed plugin: Biomedical Imaging Group http://bigwww.epfl.ch/sage/soft/watershed/
A B AB Think the pixels as particles • Think each pixel is a particle with its mass and velocity. • mAB = mA + mB • conservation of momentum • mAvA + mBvB = (mA+mB)vAB • vAB = (mAvA + mBvB)/(mA+mB) • Interpretation of mass and velocity
Think the pixels as particles (2) • Average Mass and momentum • Weighted by their mass. • The overall goal is to • bring down the effect of noise and • accelerate the process.
The algorithm • Initialize the mass and speed. • Repeat • Move particles at their speed and direction • Once two particles collide together, merge these two particles and recalculate their speed and mass. • Adjust the speed and mass according to its neighbors. • Record their paths • Until some terminate condition • Segment the image according to paths
Parameters and Options • Initial parameters • Mass • Gradient + ? • Speed • Gradient w/ tangent direction • Markers • Terminate condition • Limited Steps • Sand-box • Compete • sigma
Problems and workaround • Global color changes • Normalize the marginal distribution. • Big blank area • Use different initial mass value
Question/ Suggestion