A study on Segmentation of Remote Sensed Image using Self-Organization Map

• Main Authors: Shih-Chun Fan-Chiang, 范姜士均
• Other Authors: 吳俊霖
• Format: Others
• Language: en_US
• Published: 2006
• Online Access: http://ndltd.ncl.edu.tw/handle/48251081404710636681

碩士 === 國立中興大學 === 資訊科學系所 === 94 === An image segmentation system based on neural network is proposed for the segmentation of the color and remotely sensed images. This system is facilitated by Kohonen self-organizing map (SOM), it performs the unsupervised segmentation. In the input layer, traditional SOM only use the pixel values of R, G, and B channels, but it does not consider the relationship existed in the neighborhoods. However, in natural images, the pixels usually have strong correlation with their neighborhoods. For example, the adjacent pixels in an object are generally dependent on each other. Therefore, we propose a modified self-organizing map system is proposed in this thesis; it uses the additional spatial features in the input layer, such as the mean, medium filter, and the discrete cosine transform. In this way, we can consider both pixels themselves and their neighborhood information at the same time. And we also add a new weighting function for each neuron, which can help each neuron to be able to map to a suitable output neuron. Finally, we use application of the noise-filter to improve segmentation quality at the post-processing stage. Experimental results show that the proposed method can separate successfully the different color texture in the remotely sensed images.