Search

Image itself information is naturally robust to image reconstruction, yet most current super-resolution methods do not fully utilize global feature information. This study proposes a new image super-resolution model mixing multiple and multi-scale attentions, including two new modules: Multi-scale hybrid non-local attention upsampling module and residual dense attention block. Different from previous nonlocal methods, multi-scale hybrid non-local attention upsampling module mixes pixel-based and patch-based nonlocal attention and establishes patch-level upsampling mapping relationships at multiple scales, which enables a wider global search space. The residual dense attention block establishes attention associations in channel and spatial dimensions, which enhances the transfer and fusion of front-to-back attention information through dense connections. In this study, quantitative and qualitative evaluations are conducted on several benchmark datasets, and the experimental results show that the model outperforms similar super-resolution models in terms of performance and reconstruction quality. [ABSTRACT FROM AUTHOR]

At present, the crack segmentation model based on deep neural network has problems such as many training parameters, rough edge segmentation of cracks, insufficient segmentation accuracy, and lack of deep feature semantic information. In order to solve the above problems, the DeepLabv3 + model with better segmentation performance was studied, the Non-local attention mechanism was embedded and the backbone network ResNet101 was improved to obtain the optimized model DeepLabv3 + (N-S), and finally based on the output of the optimized model, the crack skeleton extraction method was used to quantify the crack feature parameters. The data set used was the self-made concrete beam crack image data set, and the comparison experiments of the model before and after the optimization were carried out to analyze the effectiveness of the optimization of the model in various performances, and the measured data were used to verify the quantification method for evaluating various characteristic parameters of fractures. The experimental results show that the mean pixel accuracy(mPA) and mean intersection over union( mIoU) of the DeepLabv3 + (N-S) network on the dataset have reached 88. 86% and 82. 04%, which are 2. 21% and 2. 54% higher than the original model, respectively. The crack segmentation performance is better than the original model, and the average error of quantification of various characteristic parameters of crack samples is + 8. 7%, which is lower than the original model, and can meet the requirements of engineering detection accuracy. [ABSTRACT FROM AUTHOR]

The image dehazing method based on deep learning has achieved better results than the traditional methods, and has been used widely. In this paper, image prior information is integrated into convolutional neural network, and a multi-scale image dehazing network based on non-local and prior constrain ts is proposed. In this network, a non-local and multi-scale reconstruction module is designed to capture the image s elf-similarity and reconstruction information at different scales. A loss function which contains dark channel prior, mean square error and content difference is proposed. At the same time, a phased optimization method is proposed to effectively improve the training efficiency. Compared with other dehazing methods, the proposed method can achieve better dehazing images. [ABSTRACT FROM AUTHOR]

This paper proposed a novel stereo matching method of the mobile robot based on segment tree to improve the speed and accuracy of depth recovery at the low texture, repeated texture and reflective environment. Firstly, it preprocessed the images for the mobile robot environment to reduce the influence of the light changes. In addition, this paper used the minimum spanning tree to weight the image pixels,and segmented the image by the segment tree. Then,it restored the depth of the occlusion area and the edge area that did not provide depth, and improved the accuracy of depth recovery. Finally, it carried out the comparison experiments. In indoor scenes, the calculation speed of this paper method was 187.140% faster than the ST-2 method, and semantic information could be provided for the mobile robot. On the Middlebury datasets, the calculation speed of this method was 157.500% faster than the ST-2 method, while the matching accuracy was improved 66. 547%. [ABSTRACT FROM AUTHOR]

This paper proposed an image denoising model based on structure tensor and non-local total variation, with regard to texture and structure persevering while denoising the image. This method modeled every pixel with non-local information, defined the similarity weight with both non-local information and structure tensor. The gray similarity between patched through the whole spatial domain was measured to preserve texture features; on the other hand, it took geometric characteristics of the image into account, and adjusted the weight between different contrast patches, especially increasing the weight of the low contrast region and decreasing the staireasing effect in the same time. Numerical experiments validate that the method can effectively denoise the image and preserve texture and geometry information well. [ABSTRACT FROM AUTHOR]

Copyright of Engineering Mechanics / Gongcheng Lixue is the property of Engineering Mechanics Editorial Department and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Based on the ratio distance pixel-relativity and thresholding pixel-similarity, a modified non-local filter is proposed for SAR image despeckling in this paper. Firstly, the ratio distance pixel-relativity is obtained by transforming the joint probability density function of two pixels. Then, a table of pixel-similarity threshold, as a function of the SAR image look number and neighboring reflectivity ratio, is trained according to the minimum error probability. Finally, the pixel-similarity threshold is applied to select similar pixels from the searching window for the real reflectivity estimation. The proposed approach was verified by synthetic and real SAR images, and was compared with the PPB and LHRS-PRM filters. The visual quality and the quantification comparison show that the proposed approach is excellent not only in the reconstruction of the uniform area, the character of edges, texture, and details, but also with the lower computation complexity.

A new non-local filter based on the sublook data correlation is proposed for the speckle suppression of multilook polarimetric synthetic aperture radar (PolSAR). Non-local techniques search for patches similar to the reference patch in a spatially extended neighborhood, and process the image with weighted averaging method based on the similarity. Therefore, similarity measurement plays an important role for non-local approaches.The sublook data of multilook PolSAR samples contain more correlation information than just the samples themselves, which could benefit the effectiveness of the similarity measurement in non-local filters. The proposed filter measures the correlation of two groups of sublook data in order to find the patch-based similarity. Then the homogeneous samples are selected to suppress the speckle based on the similarity. Experimental results and comparison with other filters illustrate an outstanding performance of the proposed filter. [ABSTRACT FROM AUTHOR]