Search

Face images distributed widely on social networks are vulnerable to inferring sensitive information by unauthorized automatic identification systems, which poses a threat to user privacy. To protect face privacy, several methods have been proposed to generate highly transferable adversarial faces to remove identity information. However, the results generated by existing methods still suffer from obvious perturbations that make visual perception poor, which is not friendly for sharing on social networks. This paper proposes an adversarial face generation scheme via beautification, i.e., Adv-beauty. Adv-beauty utilizes a face matcher and a beautification discriminator to collaboratively supervise the training process of the generator, prompting the generator to produce a beauty-like perturbation on the original face to confront the face matcher. In other words, the pixel changes produced by the beauty mask the undesirable visual effects produced by the perturbations. In addition, this paper sets an adversarial threshold for identity loss to prevent face distortion due to excessive deviation of identity features. Sufficient experiments show that Adv-beauty maintains good visual results and is effectively against unknown face recognition classifiers and commercial APIs.

Human action recognition and action evaluation are hot research issues in recent years. Technologies of action recognition and action evaluation share similarities in terms of data sources, data pre-processing, feature description, etc. In conjunction with the significant recent growth of application requirements, numerous studies on action recognition and evaluation appear. However, the differences and connections between human action reco-gnition and evaluation, as well as their theoretical methods and technical routes, have not been systematically analyzed and summarized. Starting from the perspective of application purpose and technical characteristics, the relationship between the two is discussed, and a clearer concept definition is given. On this basis, action recognition and action evaluation are summarized into a unified technical framework from the perspective of data processing flow. Based on this framework, the research progress and existing problems of all important links involved in motion recognition and evaluation, including data types, pre-processing, feature description, classification methods and evaluation methods, are systematically described. Among them, in the classification method section, the current action recognition classification methods are divided into statistical model-based methods and deep learning-based methods for discussion; and in the evaluation method section, based on the intervention method of expert know-ledge, the current action evaluation related work is divided into four categories and systematically sorted out. Finally, the bottlenecks and the focus of future research are summarized and prospected.

The rotation direction and speed of each element in the planetary gear mechanism can be directly shown by drawing the linear velocity vector diagram of the gear ring,planet carrier and sun wheel in the CR-CR planetary gear mechanism. The transmission ratio of typical transmission scheme is calculated by geometric method based on the obtained line speed vector diagram,which is exactly the same with the transmission ratio calculated by analytical method. The correctness and feasibility of analyzing the transmission scheme of CR-CR planetary gear mechanism by line velocity vector method is verified.

Image recoloring is an emerging editing technique that can change the color style of an image by modifying pixel values. With the rapid proliferation of social networks and image editing techniques, recolored images have seriously hampered the authenticity of the communicated information. However, there are few works specifically designed for image recoloring. Existing recoloring detection methods still have much improvement space in conventional recoloring scenarios and are ineffective in dealing with hand-crafted recolored images. For this purpose, a recolored image detection method based on inter-channel correlation was proposed for conventional recoloring and hand-crafted recoloring scenarios. Based on the phenomenon that there were significant disparities between camera imaging and recolored image generation methods, the hypothesis that recoloring operations might destroy the inter-channel correlation of natural images was proposed. The numerical analysis demonstrated that the inter-channel correlation disparities can be used as an important discriminative metric to distinguish between recolored images and natural images. Based on such new prior knowledge, the proposed method obtained the inter-channel correlation feature set of the image. The feature set was extracted from the channel co-occurrence matrix of the first-order differential residuals of the differential image. In addition, three detection scenarios were assumed based on practical situations, including scenarios with matching and mismatching between training-testing data, and scenario with hand-crafted recoloring. Experimental results show that the proposed method can accurately identify recolored images and outperforms existing methods in all three hypothetical scenarios, achieving state-of-the-art detection accuracy. In addition, the proposed method is less dependent on the amount of training data and can achieve fairly accurate prediction results with limited training data. [ABSTRACT FROM AUTHOR]

This paper approached the feasibility of quantitatively monitoring Dendrolimus spp. damage by using NOAA/ AVHRR data. The damaged rate of needle leaf was used to represent Dendrolimus spp. harming degree, and < 30%, 30%-60% and > 60% of damaged rate was defined as low, medium and severe harming degree, respectively. The correlation equation of damaged rate and normalized vegetation index (NDVI) was established, based on the ground spectrum observation. The NDVI was 0.8823 when no damage occurred. A relative NDVI value of damaged to undamaged area was used to express the remote sensing index of low, medium and severe harming degree. The index was 1 for undamaged forest, and 0.78-1, 0.57-0.78 and < 0.57 for low, medium and severe harming degrees, respectively. The mixed pixels were separated by linear addable vertical vegetation index in the monitoring, and the quantitative monitoring and analysis was accomplished for years when the three damage degrees happened. It was shown that AVHRR data could be more available in quantitatively monitoring and analyzing serious damage, while low degree damage was difficult to distinguish by AVHRR data, due to the differences of surface properties and atmospheric influences, as well as the lower space resolution of NOAA/AVHRR. The damaged area estimated by AVHRR was 12.1%-14.3% lower than that by TM.