The Research On Underwater Image Mosaic Based On SURF Characteristics

With the development and utilization of marine source continuously, underwater detecting has become the focus of world with the economic and technological competition. For underwater operations, such as seafloor exploration, sample, underwater construction, the key is to acquire accurate date of marine environment on a large range. Currently, the technology of image enhance and recovery occupies the primary position in the area of image processing. It breaks the traditional stalemate to put forward and development of underwater image mosaic against this background, which becomes an important means for gaining panoramic information without sacrificing resolution. However, due to the complex underwater environment, serious absorbing and scattering, underwater image mosaic is confronted with some situations such as fewer feature points, slower speed of extraction and higher error rate of matching in that noise and low contrast. In response to the above problems, the SURF algorithm is applied in underwater for the first time in this thesis. In view of the underwater image, especially the images acquired in turbid water, a novel method based on SURF features for underwater image mosaic with precision and good real-time is proposed.Firstly, the ROV system which is the underwater image acquisition platform is introduced briefly. Secondly, the original images are pre-processed to improve the quality of images. On the basis of pre-processing, the thesis elaborates and analyses the critical steps in the process of mosaic, including feature points’ extraction and matching, image fusion module etc. Ultimately, the integrated mosaic algorithm is completed.Image blurring, serious noise and low contrast is the common problems for underwater image. Therefore choosing a suitable de-noising algorithm for that is the key to get more detail information. In order to improve the image definition, the contrast limited adaptive histogram equalization(CLAHE) algorithm is used to process the underwater image in the pre-processing stage. In feature extraction module, the SURF(Speed-up robust feature)feature extraction algorithm,which possesses higher precision and faster speed than the classical algorithm SIFT, is adopted. This algorithm’s steps consist of the detection of spatial extreme value, determining the direction of feature point and generating the feature descriptor (64-dimensional feature vector descriptor). Moreover, the algorithm has a higher real-time performance with rotational invariance and robustness, by using the methods of integral image, Haar wavelet, etc.After completing the feature extraction, the next important step is to match the extracted feature points so as to obtain the relationship of image overlapping section. For achieving accurate matching, an improved matching algorithm called Hessian matrix’trace (i.e. the Laplacian operator) method is proposed to eliminate the error matching in this thesis. Compared with the traditional matching method of nearest neighbor search, this approach performs more optimize. Nevertheless, a lot of error matching still exists after roughly matching. For the sake of reducing the rest of mismatching, Random Sampling Consensus (RANSAC) algorithm is used to further purification for matching pairs to increase the accuracy of the match. The reason of adopting this method is that its thought about model fitting can avoid noise and the interference of erroneous data effectively.For image mosaic technology, image fusion is the last of key steps. At this stage, the direct linear transformation method is adopted to calculate the transform relation matrix of corresponding feature points in the images. With regard to the position’s difference existing in the stage of feature mapping in the process of transformation, the bilinear interpolation method is available for perfecting the phenomenon of non-corresponding. On the basis of comprehensive comparison of numerous image fusion methods, the linear gradient fusion is employed in the image fusion module. This method can eliminate the splicing gap between images and achieve better visual effects.In the experimental stage, the used underwater images were acquired in clear water and turbid water respectively. The implementation of all algorithms in experiment, including feature extraction, feature matching and image fusion owes to the use of MATLAB software. From the result of experiments, the underwater image mosaic algorithm presented in this thesis shows higher precision and accuracy, and can satisfy the requirement of the visual for underwater image mosaic.