Research On Denoising And Enhancement Algorithms For Low Illumination Image In HD Video Surveillance

To overcome the difficulty to see "details" in surveillance, HD (high definition) video technology has been rapidly developed. HD means high resolution, there are two main definitions nowadays,720P (1280×720) and 1080P (1920×1080), which are much higher than that of the analog images (the highest is D1, i.e.720×576). With high resolution security systems, better results can be obtained both in real-time surveillance and post investigation. Of course, drawbacks need to be overcome in HD video application, for example, one of them is high noise level in low illumination environment, and therefore, de-noising and enhancement processing of low illumination images are essential.In this thesis, images obtained from IP (internet protocol) camera in security system are filtered and enhanced by different technique. With evaluation of processed results, one effective approach is selected and applied in real system to improve the performance of surveillance. This these includes following parts:(1) Analyze the characteristic of low illumination level images, and collect images in normal and low illumination condition respectively in the laboratory. The results show that the noise increases with the illumination decreasing, and its gray level distribution is also compressed to lower value area. The obtained images are also utilized for filtering, and the filtered results are compared to validate different enhancement approaches.(2) Based on the characteristic of laboratory obtained low illumination level images, enhancement, time-domain and frequency-domain filters, and wavelet processing approaches are applied. The histogram, signal-to-noise evaluation and subjective observation show that the results obtained by image filtering followed by enhancement can more highly improve the image quality than applying solely processing. Among these algorithms, grave level transformation enhancement with median filter or wavelet filter has better performance.(3) To verify the feasibility and possibility of practical application, experiments were carried out for real images obtained in security system in low illumination condition. The results show that proposed approaches can improve the quality of the image and reduce the noise, and the expected results were obtained.The results of this research show that effective filter and enhancement approach can greatly improve the recognition in images obtained in low illumination environment, and can further improve the ability of digital surveillance.