The application of deep neural networks (DNNs) has significantly advanced the binary image segmentation (BIS) task. However, DNNs have been found to be susceptible to adversarial attacks involving subtle perturbations...
详细信息
The application of deep neural networks (DNNs) has significantly advanced the binary image segmentation (BIS) task. However, DNNs have been found to be susceptible to adversarial attacks involving subtle perturbations. The existing black-box attack methods usually generate one single adversarial example for different target models, leading to poor transferability. To address this issue, this paper proposes a novel adversarial example augmentation (AEA) framework to improve the transferability of black-box attacks. Our method dedicates to generating an adversarial example set (AES) which contains a set of distinct adversarial examples. Specifically, we first employ an existing model as the surrogate model which is attacked to optimize the adversarial perturbation via maximizing the binary Cross-Entropy (BCE) loss between the prediction of the surrogate model and the pseudo label, thus producing a sequence of adversarial examples. During the optimization process, besides the BCE loss, we additionally introduce deep feature losses among different adversarial examples to fully distinguish the generated adversarial examples. In this way, we can obtain an AES that contains different adversarial examples with diverse deep features to achieve the augmentation of adversarial examples. Given the diversity of the generated adversarial examples in the AES of the surrogate model, the optimal adversarial example fora certain target model is likely contained in our generated AES. Thus, the generated AES is expected to have high-transferability. In order to find the optimal adversarial example of a specific target model in the AES, we use the query method to achieve this goal. Experimental results showcase the superiority of the proposed AEA framework for black-box attack in two representative BIS tasks including salient object detection and camouflage object detection.
K-means is a classic unsupervised learning clustering algorithm. In theory, it can work well in the field of imagesegmentation. But compared with other segmentation algorithms, this algorithm needs much more computat...
详细信息
ISBN:
(纸本)9781628418293
K-means is a classic unsupervised learning clustering algorithm. In theory, it can work well in the field of imagesegmentation. But compared with other segmentation algorithms, this algorithm needs much more computation, and segmentation speed is slow. This limits its application. With the emergence of general-purpose computing on the GPU and the release of CUDA, some scholars try to implement K-means algorithm in parallel on the GPU, and applied to imagesegmentation at the same time. They have achieved some results, but the approach they use is not completely parallel, not take full advantage of GPU's super computing power. K-means algorithm has two core steps: label and update, in current parallel realization of K-means, only labeling is parallel, update operation is still serial. In this paper, both of the two steps in K-means will be parallel to improve the degree of parallelism and accelerate this algorithm. Experimental results show that this improvement has reached a much quicker speed than the previous research.
Dendroclimatic reconstructions play a key role in contextualizing recent climate change by improving our understanding of past climate variability. The Blue Intensity (BI) measurement technique is gaining prominence a...
详细信息
Dendroclimatic reconstructions play a key role in contextualizing recent climate change by improving our understanding of past climate variability. The Blue Intensity (BI) measurement technique is gaining prominence as a more accessible alternative to X-ray densitometry for producing climatically highly-sensitive tree-ring predictors. Nevertheless, accurately representing low-frequency trends and high-frequency extremes using scannerbased BI remains a challenge due to color biases and resolution limitations. Herein we introduce several methodological advances in sample surfacing, imaging, and image processing which yield measurement series analogous to BI from ultra-high-resolution (UHR;similar to 74 700 dpi) images. Such series capture changes in tree-ring anatomical density by representing wood anatomical structure using binary (i.e., black-white) segmentation of sample images. We refer to this novel technique as binary Surface Intensity (BSI). By utilizing a UHR system and entirely eliminating color and light intensity as variables, the most substantial drawbacks of scanner BI (i.e., discoloration and resolution biases) are bypassed, resulting in more accurate representations of low-frequency climatic trends and high-frequency extremes. Comparisons of several chronologies developed with the BSI and BI techniques, including a multiparameter dataset from Bjorklund et al. (2019), showed that BSI datasets outperform BI in terms of common signal (r-bar), but also contain strong climatic signals that can exceed those obtained from BI and X-ray density, and even match density datasets based on quantitative wood anatomy. However, measurement software advancements are still required to unlock the full potential of tree-ring parameters produced using the BSI technique. Ongoing development of this new technique will not only aid the attainment of long unbiased chronologies by overcoming color biases and resolution limitations, but also holds promise for unlocking UHR analys
Shrinking bias problem is a practical limitation of graph-cut-based methods for binary image segmentation. The detail parts with thin elongated object are not well preserved in graph cut minimisation as a result. The ...
详细信息
Shrinking bias problem is a practical limitation of graph-cut-based methods for binary image segmentation. The detail parts with thin elongated object are not well preserved in graph cut minimisation as a result. The authors propose to use structure transferring to overcome the problem with an expanded L2 norm of colour difference. They also show that the structure-transferring output can be integrated into graph cuts as a counter-balance to shrinking problem. Experimental results show that the proposed alpha-cutting technique is effective and efficient in improving segmentation of thin elongated objects.
We investigate a class of variational problems that incorporate in some sense curvature information of the level lines. The functionals we consider incorporate metrics defined on the orientations of pairs of line segm...
详细信息
We investigate a class of variational problems that incorporate in some sense curvature information of the level lines. The functionals we consider incorporate metrics defined on the orientations of pairs of line segments that meet in the vertices of the level lines. We discuss two particular instances: One instance that minimizes the total number of vertices of the level lines and another instance that minimizes the total sum of the absolute exterior angles between the line segments. In case of smooth level lines, the latter corresponds to the total absolute curvature. We show that these problems can be solved approximately by means of a tractable convex relaxation in higher dimensions. In our numerical experiments we present preliminary results for imagesegmentation, image denoising and image inpainting.
暂无评论