| Summary: | 碩士 === 國立交通大學 === 電控工程研究所 === 108 === In recent years, object image recognition has been a popular issue and lots of algorithms and applications are proposed. This thesis proposes a real-time multi-object recognition system based on generic Fourier descriptor and histogram of oriented gradients algorithms. The proposed system includes three parts, pre-processing, feature extraction and object recognition. First, during the period of pre-processing, the background of an input image is removed and only the objects to be recognized retain in the image. Second, generic Fourier descriptor is adopted to extract shape features of objects, and histogram of oriented gradients is used due to its invariance of geometric and photometric transformations for object orientations. Finally, the neural network classifier designed by the back-propagation algorithm is utilized to fulfill the proposed object recognition system. From the experimental results, the system could recognize objects in real-time with high accuracy rate, which verifies the effectiveness and efficiency of the proposed real-time detection system.
|
|---|
