Automation and Computational Intelligence for Road Maintenance and Management

6Feature Extraction and Fragmentation Methods

6.1 Introduction

The key goal of applying image‐processing techniques is to extract meaningful features in order to perform classification and evaluation operations. As mentioned in Chapter 5, the next step after separating the objects (distresses for example) in the image is to convert the information into properties with the goal of summarizing and reducing the amount of data for further processing. Objects isolated from the segmentation stage need to be turned into a feature vector in order to be classified and evaluated by classification methods. Extracting features with more data transfer capability can increase the speed and efficiency of the method. Various methods have been proposed to extract the features, which in a general framework, are divided into six main categories: statistical, transitional, edge, and peripheral characteristics, moment, appearance, and texture, as shown in Figure 6.1.

6.2 Low‐Level Feature Extraction Methods

Low‐level features automatically use the original image without any extracted information (spatial information) [1, 2]. Accordingly, thresholding can be considered as a kind of low‐level feature extraction that is performed as a point operation. All of these approaches can also be used to extract high‐level features. In other words, when we find objects in images, these methods can be applied on each object separately. For example, one type of pavement failure can be identified from its overall ...