Nachricht

On Wednesday, June 1, Qinyuan Yang will present intermediate results of his master's thesis with the title:

CNN Training for Detecting Curvature Extrema of Object Contours in Real Images

Abstract:

Curvature extrema along contours and their surrounding image regions are interesting local image features. Their detection in real images usually requires the following main steps: edge detection, edge refinement, edge tracing, and curvature extrema detection. Due to the complexity of this process, a curvature extrema detector based on a Convolutional Neural Network (CNN) should be developed in this work. The objective is to use real images as inputs and to obtain score maps as outputs, which describe how likely each pixel is a keypoint. This requires two steps: generation of training dataset and building a CNN architecture. The training dataset will be based on high-quality edge images from public datasets. For the edge tracing and curvature extrema detection, specific algorithms developed in the GET Lab will be used. For the CNN, the architectures from SuperPoint, LF-Net, and Key-Net will be considered. The final architecture should achieve a good accuracy and repeatability especially when images are rotated and scaled.