Decision Trees Based Image Data Mining and Its Application on Image Segmentation

In this paper, a general mining approach based on decision trees for segmenting image data is proposed. Pixel-wise image features are extracted and transformed into a database-like table that allows existing data mining algorithms to dig out useful information. Each tuple in the table has a feature descriptor consisting of a set of feature values for a given pixel along with its label. With the feature label, we can employ the decision tree to (1) discover relationship between the attributes of pixels and their target labels, (2) build a model for image processing by using the training data set. Both experiments and theoretical analysis are performed in our research. The results show that the proposed model is very efficient and effective for image mining and image segmentation. It can also be used to develop new image processing algorithms, refine existing algorithms, or act as an effective filter.