The target of the feature detection is to locate and characterize objects in an image.

Detect features

There is only one approach implemented at the moment, as this approach is doing quite well.

Approach 1: Detect contours

In this approach all contours in the (binary) image are detected and stored in ContourFeature objects. There may be objects with holes in the image, so that there are several contours for the same object. So every ContourFeature that is inside another one is removed.

Evaluation

With a good input image, this approach works very well.

It fails however when letters or words are connected with other objects. So this problems have to be solved in the image processing.

Implementation

• ContourFeature (describes the feature)
• ContourFeatureDetector (runs the detection algorithm)

Define rules for valid features

It is possible to accept or deny features based on some rules. These rules have the base class FeatureRule:

• AreaFeatureRule
• RankingFeatureRule
• SizeFeatureRule