Adaptive Detection And Reconstruction Of Boundary Surfaces In3D Images

With rapid development of3D imaging technology, the visualization of3D data has become a hot issue. The detection and reconstruction of interested boundary sur-face is becoming more and more important. And it is usually used in the description, measurement, and analysis of an object, especially in the area of medical assistant, such as diagnosis, surgical planning and organ transplantation. Detecting boundary surface using gradient threshold is a basic method, and traditionally, a global threshold is used. However, there are some images that contain gradient magnitude in a large range, from low to high. It is difficult to use a global threshold to extract the boundary surface. Moreover, due to the complexity of boundary surface and influence of noise or image details, using a global threshold usually brings unwanted image fragments or holes. What’s more, selecting a proper global threshold is also a problem, which needs to be tested many times, and depends on experience of the operator. Detecting boundary surface by an adaptive method doesn’t need to select the threshold, instead, it calculates the suitable local thresholds automatically. And it can be used in images that contain gradient magnitude from low to high as well, with high detecting preci-sion. This paper proposed three methods on adaptive detection and reconstruction of3D images.The first method is based on vision model, referring to bionics principles. Gener-ally, we judge the result of an edge detection method by comparison of images we see in2D figure slices. Therefore, the2D figures obtained by human eyes has a great impact on the judgement. However, many detecting methods doesn’t take the vision model of human eyes into consideration, which leads to a result that is different from human understandings. When our eyes recognize an edge, it relates to both the brightness of the object and the brightness of the background. By analysing the threshold-brightness curve and Weber’s law, we can use the vision model into edge detection of3D images. This method estimates gradient thresholds automatically by the gray magnitude of the pixels. Further, based on the connectivity property of boundary surface, it can connect all the edge points together, and obtain the whole surface. This method can extract boundary surfaces from most3D images, and the result is more consistent with human understandings.The second method is a combination of information in sub-images and the vision model. There are usually noises and complicated images details in3D images, which is also visible. However, they are not preferred. But the first method doesn’t take this into consideration, and extracts everything that is visible. Therefore, its result may include some unwanted fragments. As our method is focus on the step-like boundaries, whose gradient magnitudes change gradually, we can divide the3D images into many sub-images, and assume that the gradient magnitudes wouldn’t change too much in a sub-image. During the tracing of edge-cubes, we can combine the gradient information in the related sub-image and threshold under the vision model, estimating a new threshold, to separate unwanted fragments from the boundary surface. This method can improve the result of the first method, and increase the precision of detecting.The third method is based on fixing of holes. Except for unwanted fragments, an-other difficulty of edge detection is that due to the complexity of some boundary struc-tures, some boundary surfaces extracted by gradient thresholds contain holes, missing some weak boundaries. Based on this consideration, we propose an adaptive method, which needs to detect holes and trace edge-cubes repeatedly. Owing to the connectiv-ity property of the boundary surface, we can find where the tracing procedure stops. Afterwards, an appropriate algorithm is utilized to judge whether there locates a hole. If there were, that indicates the local threshold in this sub-image is too high, and it should be lower down to detect edge-cubes around the hole. And then, trace this sub-image again with the new threshold. Repeat this process once and again, until the whole boundary surface is extracted. This method can not only adjust local gradient thresh-olds to extract boundaries, but also find the location of holes, which provides reference to other hole fixing methods.The three adaptive methods proposed in this paper have been used in many exper-iments. And the results show that these methods are correct and efficient.