An Efficient DCT template-based Object Detection Method using Phase Correlation

Abstract

In this work, we propose an efficient algorithm, which utilizes the combination of discrete cosine transform (DCT) and phase correlation (PC) for fast object detection. To test the algorithm´s classification performance and computational complexity we developed a prototype and conducted several experiments with a public available car dataset. Furthermore, we compared our experimental results to a state-of-the-art object detection method. The proposed method uses the energy compaction property of DCT and requires less number of coefficients than fast Fourier transformation (FFT)-based techniques to compute PC. The computational complexity and memory requirements are significantly reduced using this method. According to our results, the proposed algorithm outperforms the baseline method with respect to training time and classification accuracy.

In this work, we propose an efficient algorithm, which utilizes the combination of discrete cosine transform (DCT) and phase correlation (PC) for fast object detection. To test the algorithm´s classification performance and computational complexity we developed a prototype and conducted several experiments with a public available car dataset. Furthermore, we compared our experimental results to a state-of-the-art object detection method. The proposed method uses the energy compaction property of DCT and requires less number of coefficients than fast Fourier transformation (FFT)-based techniques to compute PC. The computational complexity and memory requirements are significantly reduced using this method. According to our results, the proposed algorithm outperforms the baseline method with respect to training time and classification accuracy.