Physical Sciences and Mathematics Commons^™

Interpreting And Integrating Landsat Remote Sensing Image And Geographic Information System By Fuzzy Unsupervised Clustering Algorithm, Gwotsong Philip Chen

Theses

Due to the resolution of Landsat images and the multiplicity of the terrain, it is improper to assign each pixel in an image to one of a number of land cover types by using the conventional remote sensing classification method. This is also known as the hard partition method. The concept of the fuzzy set provides the means to resolve this problem. This paper presents a two-pass-mode fuzzy unsupervised clustering algorithm.

In the first passing, the cluster mean vectors which represent the geographic attributes or the land cover types are derived. In the second passing, the concept of fuzzy set …

A Heuristical Method Of Corner Points Detection On An Image Boundary, Qiulin Li

Theses

A heuristics-based method of corner points detection on an image boundary is proposed and implemented. The method uses the sampled boundary distances to find all of the candidate corner points along the image boundary. Then the curvature characteristical value is used to measure the severity of curvature change of the candidate points. Those candidates whose curvature characteristical value is under some threshold are eliminated. The paper also proposed some mechanism to reduce the effect of noises on the boundary. Experiments show that it is an efficient method and it gives satisfactory results on some image boundaries.

The Performance Of Training Pattern Sets In A New Art-Based Neural Architecture For Image Enhancement, Fu-Chun Chang

Theses

Neural network can be applied on the image enhancement after adding another two layers into the Adaptive Resonance Theory architectures (ART 1). The analysis for selecting a nice training pattern set associate the appropriate vigilance values is the main concerns in this thesis. For a single training pattern ,the network can act as a mathematical morphology operators such as erosion , dilation, opening and closing. With more than one training patterns in the network, 16 experiments are tested and are compared to each other in order to find the best selection for doing the image enhancement work. With both the …