Line Segment Matching Fusing Local Gradient Order and Non-Local Structure Information

Line segment matching is essential for industrial applications such as scene reconstruction, pattern recognition, and VSLAM. To achieve good performance under the scene with illumination changes, we propose a line segment matching method fusing local gradient order and non-local structure information. This method begins with intensity histogram multiple averaging being utilized for adaptive partitioning. After that, the line support region is divided into several sub-regions, and the whole image is divided into a few intervals. Then the sub-regions are encoded by local gradient order, and the intervals are encoded by non-local structure information of the relationship between the sampled points and the anchor points. Finally, two histograms of the encoded vectors are, respectively, normalized and cascaded. The proposed method was tested on the public datasets and compared with previous methods, which are the line-junction-line (LJL), the mean-standard deviation line descriptor (MSLD) and the line-point invariant (LPI). Experiments show that our approach has better performance than the representative methods in various scenes. Therefore, a tentative conclusion can be drawn that this method is robust and suitable for various illumination changes scenes.

Feature-Point Matching for Aerial and Ground Images by Exploiting Line Segment-Based Local-Region Constraints

In this study, we propose a feature-point matching method that is robust to viewpoint, scale, and illumination changes between aerial and ground images, to improve matching performance. First, a 3D rendering strategy is adopted to synthesize ground-view images from the 3D mesh model reconstructed from aerial images and overcome the global geometric distortion between aerial and ground images. We do not directly match feature points between the synthesized and ground images, but extract line-segment correspondences by designing a line-segment matching method that can adapt to the local geometric deformation, holes, and blurred textures on the synthesized image. Then, on the basis of the line-segment matches, local-region correspondences are constructed, and local regions on the synthesized image are propagated back to the original aerial images. Lastly, feature-point matching is performed between the aerial and ground images with the constraints of the local-region correspondences. Experimental results demonstrate that the proposed method can obtain more correct matches and higher matching precision than state-of-the-art methods. Specifically, the proposed method increases the average number of correct matches and average matching precision of the second-best method by more than five times and 40%, respectively.

Optimizing the Segmentation of a High-Resolution Image by Using a Local Scale Parameter

Image segmentation is a critical procedure in object-based identification and classification of remote sensing data. However, optimal scale-parameter selection presents a challenge, given the presence of complex landscapes and uncertain feature changes. This study proposes a local optimal segmentation approach that considers both intersegment heterogeneity and intrasegment homogeneity, uses the standard deviation and local Moran's index to explore each optimal segment across different scale parameters, and combines the optimal segments into a single layer. The optimal segment is measured by using high-spatial-resolution images. Results show that our approach out-performs and generates less error than the global optimal segmentation approach. The variety of land cover types or intrasegment homogeneity leads to segment matching with the geo-objects on different scales. Local optimal segmentation demonstrates sensitivity to land cover discrepancy and provides good performance on cross-scale segmentation.

Wavelet Based Image Coding via Image Component Prediction Using Neural Networks

In the process of wavelet based image coding, it is possible to enhance the performance by applying prediction. However, it is difficult to apply the prediction using a decoded image to the 2D DWT which is used in JPEG2000 because the decoded pixels are apart from pixels which should be predicted. Therefore, not images but DWT coefficients have been predicted. To solve this problem, predictive coding is applied for one-dimensional transform part in 2D DWT. Zhou and Yamashita proposed to use half-pixel line segment matching for the prediction of wavelet based image coding with prediction. In this research, convolutional neural networks are used as the predictor which estimates a pair of target pixels from the values of pixels which have already been decoded and adjacent to the target row. It helps to reduce the redundancy by sending the error between the real value and its predicted value. We also show its advantage by experimental results.