Object Recognition In Both Scanned Topographic Maps And Remote Sensing Images

Nowadays scanned map vectorization and intelligent processing of remote sensing images is an active area attended widely in the fields of science research and application. They play a very important role for the creation and application of digital geographic information. This thesis focuses on the automatic extraction, recognition and classification of the main objects in the scanned topographic maps and remote sensing images.In the scanned map vectorization, the thesis mainly studies the automatic extraction of the contour lines and residential area symbols in the chromatic scanned topographic maps. (1) A color key set technique based color separation modified algorithm of chromatic maps is developed, which can restrain color distortion in the scanned maps. (2) A novel method for extracting contour lines from common-conditioned topographic maps is presented, which solves contour line disconnection and conglutination by both contour line color segmentation results repaired with the line gray segmentation and local window segmentation. (3) A method to recognize the residential area symbols in the chromatic scanned topographic maps is offered, which firstly detects hatched lines by the Gabor filter and then extracts the contours of residential area symbols. (4) The topographic map vectorization software—AutoVector is developed and has been used initially.In the intelligent processing of remote sensing images, the thesis mainly studies texture feature extraction, object detection and classification in both high resolution panchromatic and multispectral images. The research of texture feature extraction contains: (1) a method called ICAG is proposed by combining Gabor wavelets and independent component analysis (ICA) for extracting texture features in high resolution panchromatic images. This method constructs two texture features: ICAG I and ICAG II, which can be used to characterize the higher-order statistical properties of the texture in different scales and orientations. (2) An ICA based multiscale texture operator is presented, which firstly makes use of multiple bands data to construct the vectors with high dimensionality and then extracts features by ICA from these vectors for multispectral texture analysis. The research of object detection and classification contains: (1) a method for recognizing towns and villages in high resolution panchromatic images is presented by fusing area segmentation based on the ICAG II texture features and edge detection. (2) A layered classification method for saline soils classification in multispectral images is proposed by combining the spectral feature, the geometric feature and the ICA based multiscale texture operator. This method solves the problems, such as the edges among saline soils and other non-saline soils are difficult to be distinguished, bare vegetation saline soils have similar spectra to that of populated areas and roads, and the measured spectra of vegetation covered saline soils are not identical.