Course Home
Administration
Syllabus
References
Lecture Notes
Readings
Assignments
Projects
Resources


Comments? Idea?























































































































































































































































































































































Computer Vision as an Engineering Problem
A Hierarchical Layer Model

Final project by

Amit Benbassat

amitbenb@cs.bgu.ac.il


Introduction

In the past decades, in spite of the efforts of many talented researches, computer vision research is not showing the hoped progress. It has dawned on me that this may in part be due to bad initial design. In this essay I set out to discover the failings in the basic design of the problem and offer a blueprint for a new design.

Approach and Method

First I covered some background work. I went over what I know about design, mainly from my study of software design and compter networks. I then examined the definition of computer vision and some of the work done in that field. I analyzed the design errors and devised a new model based on a new definition with two main modules. Focusing on one of this modules I offer to use a layer method hierachically design based on the principle of locality to further subdivide it.

Results

The first result I can say I achieved is the redefinition and remodelling of computer vision. Instead of a messy module that is somehow supposed to infer about images we have two modules. The CVM which translates real world information to discover patterns and a mental module which figures out what those patterns actually mean (A "seer" and a "thinker" if you will).

Next is the division of the CVM into layers. Based on the principle of locality we have 4 layers:
1. The physical layer that translates the incoming light into digital information.
2. The local patterns layer that performs local patterns detection (edgels etc.)
3. The large patterns layer that is responsible for finding regularities within the local patterns (edgels into edges, many small changes into movement etc.)
4. The mental interface layer that patches up the information, cleans it up, gets rid of obvious junk and sends the processed information to the mental module in its language of thought

This layer model presents a schematic solution for the problem, though the methods, especially in the last two layers are still very lacking.

Conclusions

In this essay I attempted to examine the computer vision problem, identify the faults that are causing the many difficulties in achieving good computer vision, and offer a blueprint for a solution. The solution was based on two parts. In the first computer vision was redefined and remodelled into a scheme containing two main modules; The Computer Vision Module (CVM) and the mental module. I then focused my attention to the CVM that was further divided into layers. Although the implementation of the layer model is clearly lacking in details, and the layer design itself will probably have to be modified as research progresses, I believe it is good as an initial effort. In my opinion the design problems discussed are responsible to a significant degree for the failings in computer vision research, the redefinition and remodelling of the vision problem I suggested shows better modularity and the idea of implementing the flow of information in the CVM using a layer model based on locality is good hierarchic design. Furthermore the basic layer model design also opens the door to parallelism, which is a key element in the workings of biological vision and the brain in general.

Additional Information

References

[1] A.S. Tanenbaum. Computer Networks. New Jersey, Prentice Hall PTR 2003.

[2] R. Dawkins. Climbing Mount Improbable. London, Viking 1997.

[3] V.S. Nalwa. A Guided Tour of Computer Vision. Boston, Addison-Wesley 1993.

[4] J.A. Fodor. The language of thought. New York, Crowell 1975.

[5] S. Pinker. How the mind works. New York, Norton 1997.