Impact of Distortions on the Performance of Feature Extraction and Matching Techniques

An image feature, such as edges and interest points, provides rich information on the image content and plays an important role in the area of image processing. These correspond to local regions in the image and are fundamental in many applications in image analysis. Raw data are complex and difficult to process without extracting or selecting appropriate features in advance. Feature extraction, a data reduction technique, is the transformation of large input data into a low-dimensional feature vector. It lowers the computational cost and also helps in controlling the issue of dimensionality. There are different methods of exacting features from an image and these techniques have different domains of applications. In this paper, four widely used feature detection algorithms, Harris, SURF, FAST, and BRISK feature detection algorithms are compared in terms of accuracy and time complexity for extraction and matching of feature points correctly. For this purpose, different types of transformations are added to the original images for computing the evaluating parameters like the number of features detected, matched features, and execution time required by each algorithm. Experimental results show that SURF performs better than other feature extractions and matching algorithms in terms of accuracy and run time.