No abstract

At present, most collaborative filtering algorithms use similarity as a criterion. In order to alleviate problems of cold start and sparsity in recommender system, a Collaborative Filtering Algorithm Combined with the Singular Value Decomposition (SVD) and Trust Factors (CFSVD-TF) is presented. Further mining data features, we use the SVD to mining data features to gain the implicit Items feature space, then the items-based similarity are computed by using the improved cosine similarity. The trust factor is integrated into the similarity space to generate the computable trust model. Finally, to evaluate the proposed CFSVD-TF approach, the accuracy of the CFSVD-TF algorithm has significantly improved than the traditional CF algorithm in MovieLens datasets.

In this paper we propose a multispectral image enhancement algorithm based on illuminance-reflection imaging model and morphology operation that enables us to solve the problem of improving the multispectral degraded images. Firstly, we transform the image from RGB space to HSV color space, and the hue remains unchanged. As for the saturation component, we use the adaptive nonlinear stretching to improve the image color saturation and brightness. Secondly, according to the illuminance-reflection imaging model, we adopt the guided image filtering method to decompose the brightness into illuminance component and reflection component. Usually, the illumination component mainly determines the dynamic range of the pixels in the image, corresponding to the low frequency part of the image, reflecting the global characteristics of the image and the edge detail information of the image; the reflected component represents the intrinsic essential characteristics of the image, corresponding to the high frequency part of the image, and contains most of the local detail information of the image as well as all noise. Thirdly, we present an improved adaptive gamma function, which can dynamically adjust the illuminance component by the local distribution characteristics, and use the contrast-limited adaptive histogram equalization to correct the illuminance component. Afterwards we propose a detail-feature weighted fusion strategy. The original illumination and the two corrected illuminations are fused to obtain the final illumination component. Fourthly, we propose an improved morphological operation to denoise and enhance the details of the reflection component. Finally, the corrected illumination component and the enhanced reflection component are combined to obtain the improved brightness component. In order to verify the efficiency of the algorithm proposed in the paper, we use both subjective visual effectiveness method and quantitative parameter analysis method to measure the enhancement performance in multispectral imaging scenarios, including low illumination image, underwater image, high-dynamic range image, sandstorm image, haze image and thermal infrared image. Then standard deviation, information entropy and average gradient are used as evaluation indices respectively, and qualitative and quantitative comparison with a variety of image enhancement algorithms show that the proposed algorithm can not only well suppress noise but also obviously improve local details and global contrast. Experimental results show that the proposed method proves to be better in performance.

Users' feedback information as the ground-truth has attracted a lot of attention in recommender systems. However, the feedback that could be contaminated by users' misoperations or malicious operations is probably not true in real scenarios. This work aims to develop a technique based on an improved Bayesian personalized ranking (BPR), called adversarial training-based mean Bayesian personalized ranking (AT-MBPR). In this method, we divide the feedback information into three categories based on the mean Bayesian personalized ranking (MBPR), then gain the implicit feedback from the mean and non-observed items of each user, following which, adversarial perturbations are added on the embedding vectors of the users and items by playing a minimax game to reduce the noise. The experiments demonstrate in five datasets that our approach outperforms the traditional BPR methods and state-of-the-art methods used for the recommendation. Our implementation is available at: https://github.com/ HanXia001/ Adversarial -Training-based-Mean-BPR-for-Recommender.

No abstract