Research On Underwater Image Ehancement And Object Recognition Algorithms

The development of digital image technolody is relatively mature, and it has achieved remarkable results in many fields, while the study of underwater image processing techniques is rarely effective. The environment of underwater is much more complex than on land, and as there is no light underwater, underwater imaging systems have to rely on the artificial light to provide illumination. Also, when light transmits in the water, it will suffer serious attenuation due to the water absorption, reflection and scattering. The images we are interested on can suffer from one or more of the following problems:limited range visibility, low contrast, non uniform lighting, blurring, bright artifacts, color diminished (bluish appearance) and noise. The purpose of image enhancement aims to improve or solve one or some of the above problems; therefore, underwater image enhancement technology has become one of the key research content of underwater image processing technology.Underwater underwater object recognition technology is based on image processing technology, using optical images for underwater object recognition has become another important topic in the field of digital image processing. At present, although the underwater object recognition technologies based on sonar image play a dominant role in underwater vision applications, due to optical image can provide more intuitive and rich information. It has irreplaceable advantages comparing with the acoustic technology especially in celar water or in short range operation of underwater vehcile. For complex underwater environment, how to extract interesting objects rapidly and exactly is the key of underwater object recognition.In the dissertition, we mainly focus on analysising and studying the theories and methods of underwater image enhancement technology and object recognition algorithms. The mainly research of our work can be summarized as follows:(1) Research of color space selection for underwater image enhancement and object recognitionThe basic theory of color space and dichromatic reflection model are presented and their characteristics are discussed, respectively. In the application of underwater image enhancement, three traditional classic image enhancement algorithms that are contrast limited adaptive histogram equalization (CLAHE), homomorphic filter (HF) and wavelet threshold denoising are firstly described. Their performance for underwater image enhancement in six classic color spaces is analysed in detail at the experiment segment. In the application of underwater object recognition based on color feature, The six hybrid color space built with r, g, b, H, S, y, u, v, c1, c2, c3 component are defined and analyzed based on dichromatic reflection model. The experiments are mainly focusing on the influence on the correct recognition rate under different underwater situations, for example illumination intensity, image captured angle and the distance from camera to object. These real data and conclusions can provide an experimental basis for the optimal choice of color space, and also can be regards as reference examples for the following research work.(2) Research of underwater image enhancement method based on hue preservingAfter analysis and comparison of the shortcomings of existing methods for underwater image enhancement, an underwater image enhancement method based on hue preserving is proposed by combining with HSI and HSV color spaces. Firstly, the image is converted from RGB space to HSI space, the hue component H is preserved and homomorphic filtering is applied on the saturation component of S and intensity component I. Then converting it to HSV color space, the component H is preserve constantly as well and histogram stretching is applied on the S and V component. Finally, the noise amplified by the previous two operations is suppressed using wavelet-domain denoising based on threshold processing. After comparing with several other algorithms, it can be conclude that our proposed approach gives better results especially in improving the non-uniform illumination and balancing the color and contrast.(3) Research of underwater image object recognition based on color and shape featuresDue to the complexity of the underwater environment, the feasibility and advantage of underwater object extraction and recognition based on color and shape features are dicussed. The principle of underwater object extraction is analyzed according to color modification model. Relying on the estimation of the attenuation parameters and prior color, all the "compatible colors" of underwater object can be detected. Since the close "compatible colors" may be existing in the background region, which will lead to false detection. Faced with the above problems, Erosion operation in binary morphology can be used to remove small pieces and deburring of object boundaries, conneted domain detection algorithm can help eliminate the larger background noise, and expansion operation plays a role in padding the object internal holes and smoothing object boundaries. After accurately extract the interest object, the canny edge detection operator is applied to obtain object edge contour, and shape signature based method is proposed for shape recognition. The experimental results demonstrate that our methods have good ability to underwater object detection and recognition in complex underwater scene. By comparing with the other two algorithms, our proposed approach outperforms them especially in terms of the accuracy and real time.