2020/04/14 by Kelm, André Peter, Udo Zölzer, Zölzer, Udo · 1 citation
Computer Science · #Medical Image Segmentation Techniques #Advanced Neural Network Applications #Image and Object Detection Techniques
paper · pdf · doi:10.48550/arxiv.2004.06587
We develop a new contour tracing algorithm to enhance the results of the\nlatest object contour detectors. The goal is to achieve a perfectly closed, 1\npixel wide and detailed object contour, since this type of contour could be\nanalyzed using methods such as Fourier descriptors. Convolutional Neural\nNetworks (CNNs) are rarely used for contour tracing. However, we find CNNs are\ntailor-made for this task and that's why we present the Walk the Lines (WtL)\nalgorithm, a standard regression CNN trained to follow object contours. To make\nthe first step, we train the CNN only on ship contours, but the principle is\nalso applicable to other objects. Input data are the image and the associated\nobject contour prediction of the recently published RefineContourNet. The WtL\ngets a center pixel, which defines an input section and an angle for rotating\nthis section. Ideally, the center pixel moves on the contour, while the angle\ndescribes upcoming directional contour changes. The WtL predicts its steps\npixelwise in a selfrouting way. To obtain a complete object contour the WtL\nruns in parallel at different image locations and the traces of its individual\npaths are summed. In contrast to the comparable Non-Maximum Suppression method,\nour approach produces connected contours with finer details. Finally, the\nobject contour is binarized under the condition of being closed. In case all\nprocedures work as desired, excellent ship segmentations with high IoUs are\nproduced, showing details such as antennas and ship superstructures that are\neasily omitted by other segmentation methods.\n