Search

For compressed images and videos, quality enhancement is essential. Though there have been remarkable achievements related to deep learning, deep learning models are too large to apply to real-time tasks. Therefore, a fast multi-frame quality enhancement method for compressed video, named Fast-MFQE, is proposed to meet the requirement of video-quality enhancement for real-time applications. There are three main modules in this method. One is the image pre-processing building module (IPPB), which is used to reduce redundant information of input images. The second one is the spatio-temporal fusion attention (STFA) module. It is introduced to effectively merge temporal and spatial information of input video frames. The third one is the feature reconstruction network (FRN), which is developed to effectively reconstruct and enhance the spatio-temporal information. Experimental results demonstrate that the proposed method outperforms state-of-the-art methods in terms of lightweight parameters, inference speed, and quality enhancement performance. Even at a resolution of 1080p, the Fast-MFQE achieves a remarkable inference speed of over 25 frames per second, while providing a PSNR increase of 19.6% on average when QP = 37.

Abstract Vehicle re-identification is a challenging task that matches vehicle images captured by different cameras. Recent vehicle re-identification approaches exploit complex deep networks to learn viewpoint robust features for obtaining accurate re-identification results, which causes large computations in their testing phases to restrict the vehicle re-identification speed. In this paper, we propose a viewpoint robust knowledge distillation (VRKD) method for accelerating vehicle re-identification. The VRKD method consists of a complex teacher network and a simple student network. Specifically, the teacher network uses quadruple directional deep networks to learn viewpoint robust features. The student network only contains a shallow backbone sub-network and a global average pooling layer. The student network distills viewpoint robust knowledge from the teacher network via minimizing the Kullback-Leibler divergence between the posterior probability distributions resulted from the student and teacher networks. As a result, the vehicle re-identification speed is significantly accelerated since only the student network of small testing computations is demanded. Experiments on VeRi776 and VehicleID datasets show that the proposed VRKD method outperforms many state-of-the-art vehicle re-identification approaches with better accurate and speed performance.

In this paper, a fast inter mode decision algorithm, called the unimodal model-based inter mode decision (UMIMD), is proposed for the latest video coding standard, the high-efficiency video coding. Through extensive simulations, it has been observed that a unimodal model (i.e., with only one global minimum value) can be established among the size of different prediction unit (PU) modes and their resulted rate-distortion (RD) costs for each quad-tree partitioned coding tree unit (CTU). To guarantee the unimodality and further search the optimal operating point over this function for each CTU, all the PU modes need to be first classified into 11 mode classes according to their sizes. These classes are then properly ordered and sequentially checked according to the class index, from small to large so that the optimal mode can be early identified by checking when the RD cost starts to arise. In addition, an effective instant SKIP mode termination scheme is developed by simply checking the SKIP mode against a pre-determined threshold to further reduce the computational complexity. The extensive simulation results have shown that the proposed UMIMD algorithm is able to individually achieve a significant reduction on computational complexity at the encoder by 61.9% and 64.2% on average while incurring only 1.7% and 2.1% increment on the total Bjontegaard delta bit rate (BDBR) for the low delay and random access test conditions, compared with the exhaustive mode decision in the HEVC. Moreover, the experimental results have further demonstrated that the proposed UMIMD algorithm outperforms multiple state-of-the-art methods.

3D-HEVC explores several new intra-frame depth coding tools to improve the coding efficiency, but at the price of high computational complexity, which hinders the practical applications of 3D-HEVC. Therefore, a bi-layer texture discriminant fast depth intra coding for 3D-HEVC is proposed in this paper. With a sum of gradient matrix (SGM), the texture complexity (TC) of current depth coding unit (CU) and its subblocks are calculated at once. The values of TCs indicate whether to split the depth CU further and skip unnecessary DMMs checking. The experimental results show that compared with the original 3D-HEVC, the proposed algorithm reduces the depth encoding time by 44.8%, and the average bit rate of the synthetic viewpoint is only increased by 0.38%, which outperforms the state-of-the-art fast 3D-HEVC algorithms.

The just noticeable difference (JND) models in pixel domain are generally composed of luminance adaptation (LA) and contrast masking (CM), which takes edge masking (EM) and texture masking (TM) into consideration. However, in existing pixel-wise JND models, CM is not evaluated appropriately since they overestimate the masking effect of regular oriented texture regions and neglect the visual attention characteristic of human eyes for the real image. In this work, a novel JND model in pixel domain is proposed, where orderly texture masking (OTM) for regular texture areas (also called orderly texture regions) and disorderly texture masking (DTM) for complex texture areas (also called disorderly texture regions) are presented based on the orientation complexity. Meanwhile, the visual saliency is set as the weighting factor and is incorporated into CM evaluation to enhance JND thresholds. Experimental results indicate that compared with existing relevant JND profiles, the proposed JND model tolerates more distortion in the same perceptual quality, and brings better visual perception in the same level of the injected JND-noise energy.

In this paper, a joint feature and similarity deep learning (JFSDL) method for vehicle reidentification is proposed. The proposed JFSDL method applies a siamese deep network to extract deep learning features for an input vehicle image pair simultaneously. The siamese deep network is learned under the joint identification and verification supervision. The joint identification and verification supervision is realized by linearly combining two softmax functions and one hybrid similarity learning function. Moreover, based on the hybrid similarity learning function, the similarity score between the input vehicle image pair is also obtained by simultaneously projecting the element-wise absolute difference and multiplication of the corresponding deep learning feature pair with a group of learned weight coefficients. Extensive experiments show that the proposed JFSDL method is superior to multiple state-of-the-art vehicle re-identification methods on both the VehicleID and VeRi data sets.