Automatic Recognition Of Objects Based On Multi-signature

The automatic recognition of the object is always one of the most interesting topics in image processing and artificial intelligence. In the processing of remote sensing image the topic is regarded as the automatic recognition of the objects on the earth. It is very difficult to recognize the differences among tree species automatically even on a high-resolution remotely sensed image. Instead of relying on only such a single signature as color, texture or edge shape , a multi-signature analyzing method has been used in the thesis.Several color infrared aerial images in the scale of 1:5000 were taken as the data resources and many existed algorithms of extracting signatures on a remotely sensed image have been tested, compared, estimated and filtered for recognizing those differences in tree species. There are some of those tested algorithms as the histogram character vectors, color moments and standard error of the region's color value based on color; the gray co-occurrence matrix, gray gradient co-occurrence matrix, auto-correlation function model, fractal texture model, gray primitive co-occurrence matrix, wavelet transformation, Fourier transformation energy spectrum based on texture; the edge moments, edge direction histogram, circular degree and rectangular degree based on the edge shape of tree species.Then the classification results are analyzed and some new advices on different algorithms are proposed. The opinion that only use one signature, no matter color or texture or edge shape to recognize the unknown objects, would get no accurate result is proposed in the thesis too.Furthermore, the developing aspects of the multi-signature recognition in the processing of remote sensing image and the problems may appear when multi-signature recognition is used in practice are discussed in the thesis too.