Learning effective feature representations and similarity measures are crucial to the retrieval performance of a content-based image (CBIR) system. Despite extensive research efforts for decades, it remains one most challenging open problems that considerably hinders successes real-world CBIR systems. The key challenge has been attributed well-known ``semantic gap'' issue exists between low-level pixels captured by machines high-level semantic concepts perceived human. Among various...

Solving long-tail large vocabulary object detection with deep learning based models is a challenging and demanding task, which however under-explored. In this work, we provide the first systematic analysis on underperformance of state-of-the-art in front distribution. We find existing methods are unable to model few-shot classes when dataset extremely skewed, can result classifier imbalance terms parameter magnitude. Directly adapting classification frameworks not solve problem due intrinsic...

The problem of recognizing actions in realistic videos is challenging yet absorbing owing to its great potentials many practical applications. Most previous research limited due the use simplified action databases under controlled environments or focus on excessively localized features without sufficiently encapsulating spatio-temporal context. In this paper, we propose model context information a hierarchical way, where three levels are exploited ascending order abstraction: 1) point-level...

State-of-the-art near-duplicate image search systems mostly build on the bag-of-local features (BOF) representation. While favorable for simplicity and scalability, these have three shortcomings: 1) high time complexity of local feature detection; 2) discriminability reduction descriptors due to BOF quantization; 3) neglect geometric relationships among after To overcome shortcomings, we propose a novel framework by using graphics processing units (GPU). The main contributions our method...

Resistance spot welding poses potential challenges for automotive manufacturing enterprises with regard to ensuring the real-time and accurate quality detection of each spot. Nowadays, many machine learning deep methods have been proposed utilize monitored sensor data solve these challenges. However, poor results or process interpretations are still unaddressed key issues. To bridge gap, this paper takes bodies as objects, proposes a resistance online method dynamic current based on combined...

In this paper, we proposed parallel and pipeline architecture for the sub-pixel interpolation filter in H.264/AVC conformed HDTV decoder. To efficiently use bus bandwidth, bring forward three memory access optimization strategies to avoid redundant data transfer improve utilization. processing throughput, multi-stage conducting transmission filtering parallel. Moreover, balance tradeoff between accessing scheme granularity devise a dedicated buffer organization convert tree-structured block...

This paper points out some defects in the techniques used H.264 rate control and presents several new algorithms to improve them. The improved algorithm has following main features: 1) bits allocated each P-frame is proportional local motion it, i.e, more are a frame if it stronger; 2) quantization parameter for I-frame choosed based on allocation scheme I-frame; 3)the calculation simple encoding complexity prediction scheme, which robust of less than quadratic model by low bit video coding....

Both the motion-detection and infra-field interpolation filter are important factors affect efficiency of motion adaptive de-interlacing. New accurate detection (AMD) algorithm is proposed to improve accuracy detection, which reduces possibility error with a median filter. To intra-field deinterlacing in moving regions, an anti-aliasing (AAIF) proposed, better than typical windowed sine function. The simulation results show that peak signal noise ratio (PSNR) our method 0.5-7.5 dB higher...

This paper proposes a high efficiency memory controller for an H.264 HDTV decoder with synchronous DRAMs. As adopts tree structured (supports small block size) motion compensation, the bandwidth requirement of is higher than previous video processing algorithms. requires to be optimized. Based on decoding data access behavior analysis, SDRAM new mapping method has been designed reduce overhead cycles page-activation. Experiment results indicate that improved by one-third performance bus...

Scrambling is widely used to protect privacy in surveillance video. However, as a critical issue protected video scrambling, drift error has not been adequately studied. In this paper, we focus on prevention for different elements scrambling H.264/AVC video, which the prevailing coding standard. A restricted scheme proposed prevent Transform Coefficient (TC), Intra Prediction Mode (IPM) and Motion Vector (MV) respectively. Experiments show that effectively prevents with efficiency...

Multivariate time series prediction is a significant research area that aims to forecast future values based on past observations. Deep learning models with attention mechanisms have shown good predictive performance by emphasizing optimal-related sequences in the target series. However, these ignore mutation information of nontarget and long short-term dependencies. To this end, divide-and-rule combined method proposed address limitations, which uses differentiated feature extractors...

In this paper, we propose a motion adaptive deinterlacing method with texture detection for scan-rate conversion of video data. The basic idea is to classify the missing pixel four different regions including smooth region, static and region based on results detection, then interpolation methods are used produce pixel. Extensive simulations conducted sequences indicate that performance proposed superior previous direct merging method, bilinear interpolation, ELA, vertical temporal median...