Remote-sensing Image Processing and Recognition

The study topics of multispectral remote-sensing image processing, analysis, and recognition include: texture segmentation, image fusion, subpixel analysis, feature extraction, and land feature matching. We used the techniques of Markov random field and genetic algorithm to segment the multispectral texture images. We utilized a wavelet-based method to fuse a low-resolution multispectral image and a high-resolution single-spectral image into a high-resolution multispectral image. We applied the technique of linear mixed model and the singular value decomposition to analyze the material components of image pixels. We used  the discrete wavelet transform and hidden Markov model to enhance the quality of images. We used the method based on the discrete wavelet transform and line tracking to extract edges; such as, road networks, rivers, lake contours, coastal lines, etc. We match land cover with the extracted texture regions and the edge/line features. Moreover, those features will also be used for the detection of land cover change, target detection on the ground, area estimation, and the synthesis of digital feature analysis data for the land usage planning andother applications.


Results: Texture segmentation, Image fusion, Subpixel analysis, Feature extraction,

land cover matching


Members: C.-H. Lou, C.-T. Tou, I.-C. Deng

(C.-C. Lai, I.-W. Tsai, H.-M. Tsai, C.-R. Yang, G.-C. Lin, I.-L. Chen)


  Texture segmentation  

Examples of texture segmentation on synthesis color images

    Original images.

    Segmented results.



Examples of texture segmentation on SPOT images.

     Original images.

      Segmented results.

Multispectral Image Fusion 


(a) Original image.                 (b) Wavelet-based PCA.                    (c) Standard IHS.


(d) Traditional PCA.                  (e) Selecting max.                (f) Replacing partial coefficients.


Mixed-pixel Analysis 

Original image

Fraction images

A FLCI  image.


Red clover.

Ground truth.