Jindong Xu

Multi-scale hybrid saliency analysis for region of interest detection in very high resolution remote sensing images

Abstract Researchers have recently been performing region of interest detection in such applications as object recognition, object segmentation, and adaptive coding. In this paper, a novel region of interest detection model that is based on visually salient regions is introduced by utilizing the frequency and space domain features in very high resolution remote sensing images. First, the frequency domain features that are based on a multi-scale spectrum residual algorithm are extracted to yield intensity features. Next, we extract the color and orientation features by generating space dynamic pyramids. Then, spectral features are obtained by analyzing spectral information content. In addition, a multi-scale feature fusion method is proposed to generate a saliency map. Finally, the detected visual saliency regions are described using adaptive threshold segmentation. Compared with existing models, our model eliminates the background information effectively and highlights the salient detected results with well-defined boundaries and shapes. Moreover, an experimental evaluation indicates promising results from our model with respect to the accuracy of detection results.

A fast mixing matrix estimation method in the wavelet domain

In the field of blind image separation (BIS) based on the sparse component analysis, separation efficiency and accuracy are directly affected by the number of clustering samples. To address this problem, a new algorithm for the detection of points in the Haar wavelet domain was proposed in which only single source contributions occur. The algorithm identified the single source points (SSPs) by comparing the absolute direction between the diagonal and horizontal components of the Haar wavelet coefficients of mixed images. After screening the SSPs, the wavelet coefficients of the images are sparser. The experimental results showed that, compared to the conventional method, the proposed algorithm could estimate the mixing matrix faster and more accurately, and it allowed identification of latent variables via statistical histograms.

Scale-invariant feature transform based on the frequency spectrum and the grid for remote sensing image registration

A novel image registration method is proposed based on the scale-invariant feature transform (SIFT). Our approach combines the image frequency spectrum with a regular grid. The method exploits features of SIFT, the greater number of feature points in high-frequency areas of an image and regular grid traits with a fine spatial distribution to extract evenly distributed feature points. Global correspondence error checking is used to remove false matching pairs. The registration of eight groups of remote sensing images shows that our method has well-distributed feature points in the spatial and scale domains, obtains an average of 15.78% matching pairs more than uniform robust (UR)-SIFT, and provides better registration accuracy than SIFT and UR-SIFT.