Course Home
Administration
Syllabus
References
Lecture Notes
Readings
Assignments
Projects
Resources


Comments? Idea?























































































































































































































































































































































multi choise test - automatic gradeing system

Final project by

Roii spoliansky

roii@bgu.ac.il


Introduction

Going over multiple choice tests is a time consuming process, that can be better handled be an automatic system automatic system in market today are very exspesive, allow very low maneuvering room for the test writer and for the student. In order to get better usage of researchers time and have no financial costs a pc and scanner based system is proposed.

The goal of the project is to correctly extract the answers from a multiple choice test.



Approach and Method

The approach taken in this project was dividing the process into steps according to the steps performed by a human system.

The steps are:

  • Align the page
  • Identification of columns and rows
  • Decision of what is included in every slot
  • Separation into parts if necessary
  • Decision which slot is the final check
  • Scoring for each question


Align the page

aligning the page was performed by detecting the edges in the scan (canny edge detection), performing convolution on the scan the connect both edge of the line to one, using radon transform with the principles of hough transform to detect the position and angle of the lines. rotating the scan to with the angle received from the radon transform.

Identification of columns and rows

Vary simalr to the first step detecting the edges in the scan (canny edge detection), performing convolution on the scan the connect both edge of the line to one, using radon transform with the principles of hough transform to detect the position and angle of the lines. the position of the lines is saved as number of rows and columns in the matrix.

Decision of what is included in every slot

after decading if the frame is part of the lines detected we sum the values of the RGB, and segment the writing from the background. for each slot we srink it size in order to make sure we have no lines in the slot.

Separation into parts if necessary

If a mark is larger the a predetermined size we check for spacing in the middle. slot with two parts or more are split in to two part according to a neighbor rule of 8 neighbors. After splitting the parts and removing noise the smallest part is chosen and the others removed.

Decision which slot is the final check

Only slots that are larger than giving threshold are considered to be the chosen slot. each slots if test for his comptebilety to the target object (x or v). if to slot are a mutch then we take the smallest one, if no slot is a mutch we take the largest one.

Scoring for each question

question scoring is being performed by multiplying the each cell in the checked matrix with each cell in the score matrix. To get the final grade we sum the Multiplied matix's value.

The output is the grades, the mean score, the stenderd diviation, an historgram and images of scans with currect and not currect marks.



The system was designed for easy use and has the following screen:



Results

The system was tested on pre prepared tables and had 100% correct classifications of marks.
The system was tested on a class test with 28 students and had 100& correct classifications of marks.

Conclusions

Main conclousion - many systems today can be replaced by computer vision systems.
Many topics for other fields and for with in computer vision can be adopted and used for computer vision purposes.
Designing computer vision system process can be based on human thinking process.
The system is ready to use, but before deploying it in all university departments it is recommended to check the system on other tests and scanners.

Additional Information

References

  • Al-amri, Salem Saleh, and Namdeo V. Kalyankar. "Image segmentation by using threshold techniques." arXiv preprint arXiv:1005.4020 (2010)
  • Beylkin, Gregory. "Discrete radon transform." Acoustics, Speech and Signal Processing, IEEE Transactions on 35.2 (1987): 162-172
  • Canny, John. "A computational approach to edge detection." Pattern Analysis and Machine Intelligence, IEEE Transactions on 6 (1986): 679-698.
  • Illingworth, John, and Josef Kittler. "A survey of the Hough transform." Computer vision, graphics, and image processing 44.1 (1988): 87-116.