Boundary detection of color images using neural networks
Haruyuki Iwata, T. Agui, Hiroshi Nagahashi · 2002
In boundary detection of color images, it is essential to form local edge elements detected by a local edge detection method into groups for finding straight or curved lines. A new boundary detection method based on the Hopfield neural network is proposed. First, an image is divided into blocks. In each block, at most two edge segments are detected by a proposed edge tracing method. Then, a unit of the Hopfield neural network is assigned to each edge segment. Some properties of edge segments belonging to a boundary, such as colors and directions, are embedded in an objective function of the network, and the boundary is detected by minimizing the function. To reduce computation time, a fast algorithm of a boundary detection method is also proposed. The experimental results show that the proposed method is applicable for the partially disconnected and/or blurred boundaries.