Image Processing Basics

As noted, there are some basic image manipulation functions that must be used by any computer vision system. Here are a few of the more common functions used. These descriptions are intended to give a general idea of the concepts, and may not be technically exact.

Grayscale Conversion
Simply put, this is the process of taking a full color image, and converting it to black, white, and shades of gray. This is important in many functions because color only confuses the computer. If you want to know what some text says, it doesn't really matter what color it is. By converting to grayscale, we can eliminate the extra color information that might confuse the computer.

Binarization
This process is much like grayscale conversion, except that the resulting image contains only black and white, no gray. This step is usually taken after some other preprocessing is done. For example, in detecting the text in a picture, you would first attempt to distinguish the text from the background, then apply binarization to convert all text data to black, and all of the background to white.

Edge Detection
Edge detection is a way for the computer to detect the edges or outlines of objects. Edge detection is usually done by detecting sudden drastic changes in the color or brightness of two adjacent pixels in the image. By detecting edges, the computer can use this information to process the original images to remove all but the object of interest. That object can then be processed for whatever reason necessary.

Noise Filtering / Removal
This is another very important process. Almost any image will contain some degree of extra data. For example, in scanning a resume printed on fancy paper, the scanned image will have some type of pattern or grain that was picked up from the paper. For example:

graphitepaper.jpg (13408 bytes)   yellowpaper.jpg (10501 bytes)

The trick here is to remove the pattern or grain so that we end up with just the text. This is a difficult task to perform, but there are special algorithms which can do it nicely.

 

NEXT TOPIC - OPTICAL CHARACTER RECOGNITION

PREVIOUS TOPIC - INTRODUCTION

 

return to the beginning