Project ID:2013s03
Students:Mark Gakman, Herzel Abramov
Supervisor:Berger Israel

(Link to project documentation folder to be placed here)

Classical and Advanced methods of image sharpening and edge detecting


When taking a picture there is a blur and an additive noise that derives from optical and mechanical aspects of the camera. These flaws are often preventing us from detecting small yet important details in certain pictures. In this project we will try to improve the resolution of “Printed Circuit Board” images that are used to detect flaws on the board.  In the first part we will use the classical methods and check the improvement over the blurred image. In the second part of the project we will test new advanced methods of applying partial differential equations (iteratively) in order to improve the resolution, and compare it to the previous part.

PCB (printed circuit board) Industry Problem


High resolution image

  • –Pros  :  good flaw detection ability
  • –Cons :  slow, expensive ,memory

Low resolution image

  • –Pros  :  fast, less expensive,  less memory
  • –Cons:   degrading  the ability of flaw detection




figure 1: Example of high and low resolution

Main Objective

—To bring the low-resolution image as close as possible to the high-resolution image.

Work Steps

  1. —Checkerboard
    • Simulate a camera operation on high resolution image
    • Exploring methods of image reconstruction
    • Implementing those methods on the image
    • Analyzing the results with measuring tools
  2. —Real PCB images
    • Trying the methods on PCB images
  3. ——Error Measurement
  4. —Partial Differential Equations (Diffusion) for edge sharpening







Average PSF (Point Spread Function) variances  lead to great results: clearer and sharper borders , easily recognizable flaws.


Future Work

  • Derivative statistics:As an improvement of the edge sharpning method , we would suggest a wiser choice for the parameters.  If we look at the histogram of the image, we will expect to see several peaks and we can estimate these parameters


  • Hough transform:Thanks to our good edge results, we can use Hough transform to detect lines.
    Using these lines, one can think of great algorithms of automatic flaws detection! (for example : integration over a line, and mark low values by “suspicious zones” )