Image segmentation based on wavelet feature descriptor and dimensionality reduction applied to remote sensing

Image segmentation is a fundamental stage in several domains of knowledge, such as computer vision, medical applications, and remote sensing. Using feature descriptors based on color, pixel intensity, shape, or texture, it divides an image into regions of interest that can be further analyzed by higher level modules. This work proposes a two-stage image segmentation method that maintains an adequate discrimination of details while allowing a reduction in the computational cost. In the first stage, feature descriptors extracted using the wavelet transform are employed to describe and classify homogeneous regions in the image. Then, a classification scheme based on partial least squares is applied to those pixels not classified during the first stage. Experimental results evaluate the effectiveness of the proposed method and compares it with a segmentation approach that considers Euclidean distance instead of the partial least squares for the second stage.