Fast Template Matching Using Pruning Strategy with Haar-like Features

Template matching is one of the most popular problems in computer vision applications. Many methods are proposed to enhance the accuracy and performance in terms of processing time. With the rapid development of digital cameras/recorders and HD video, images captured by modern devices have much higher resolution than before. Therefore it is necessary to have novel real-time processing template matching algorithms. This motivates our proposal to improve the speed of matching an arbitrary given template, especially in a large size image. Our key idea is based on a pruning strategy to remove certain unmatchable positions in a large size image with only a few simple computational operations. Our proposed method has a flexibility for users to select various dissimilarity measures to increase the accuracy of the system in practical applications. Experiments show that our algorithm is faster than the standard algorithm implemented in OpenCV using Fast Fourier Transform about 8 to 10 times.