Study Of Edge Detection Theory,Fast Algorithms And Real-time Application Based On Gabor And Wavelet

Edge detection is the first step of image segmentation based on edge. Edge usually owns the main information of image and lies in the anomalous and unstable location which gives the position of image profile. The profile is usually the most important characters of image during the processing of image, so it is necessary for us to detect the image and get the edge image.The conventional methods of edge detection , such as differential operators or edge fitting operators, are not easy to achieve the tradeoff between precision and anti-noise, so the effect of edge detection is not satisfying. The paper applies the Gabor method, which can achieve joint minimal uncertainty and optimum joint spatial/spatial-frequency localization, to detect the edge. Though some persons have presented the idea to use Gabor for edge detection before, they only gave the one-dimensional(1D) Gabor filter-based edge detection method for step edge. This paper not only gives the two-dimensional (2D) Gabor filter-based edge detection method for step edge, also the one-dimensional(1D) method for ramp edge.The conventional methods of edge detection have good effect on clear image, but are sensitive to noise. Though Mallat used multiresolution wavelet to detect edge of noisy images, it was not effective for much noisy image. Multiresolution wavelet is widely used in antinoise application because of the correlation of different layer wavelet coefficients. This paper presents a method called "shifting different layer coefficients" to suppress noise and pick-up the edge. The new method is robust on noisy edge detection.Though the two new methods can improve the effect of edge detection, they need the Gabor and wavelet transforms which make the computation complex and cost much time and are not easy to used in real-time application. This paper presents an efficient fast approach to compute real discrete Gabor coefficients by applying the FHT(Fast Hartley Transform) in Gabor Transform, also the FFT (Fast Fourier Transform) is proposed to used in compute the wavelet coefficients.In order to apply new idea in realtime application, such as real-time platerecognition,not only new fast algorithms are necessary,but also fast hardware which is a key for real-time application. The hardware architecture should be compatible with the algorithms for maximizing the utilization of the available resources of the hardware. In this paper,the design and implementation of real-time image capture board based on TMS320C6205 DSP (Digital Signal Processor) for plate recognition and FFT algorithm is studied. Also,the methodology for hand-coding algorithms and software pipelines for TMS320C6205 VLIW-type processors are proposed for improving the code efficiency. Last,the optimized methods for realizing plate recognition system and FFT are applied. Experiment result shows that the board has realtime performance for plate recognition and FFT algorithm and is good for application.