Search

As perception-related applications are of great importance in industrial production and daily life nowadays, solutions for understanding given images semantically receive numerous attention from the literature. To this end, significant accomplishments have been reached for such pixel-wise segmentation problem thanks to novel manipulations of integrating global context into local details in convolutional neural networks. However, this strategy in the existing work did not exhaustively exploit middle-level features, which carry reasonable balance between fine-grained and semantic information. Therefore, this paper introduces a Cross-Attentional Bracket-shaped Convolutional Network (CAB-Net) to leverage their contribution to the tournament of constructing pixel-wise labeled map. In concrete, fine-to-coarse feature maps of interest from the backbone network are densely combined by an efficient fusion of channel-wisely and spatially attentional schemes in crossing manner, namely Cross-Attentional Fusion, to embed semantically rich features into finer patterns. Continuously, these newly decoded outputs repeat the same procedure round-by-round until shaping a final feature map having finest resolution for complete scene understanding. Consequently, the proposed CAB-Net achieves competitive mean Intersection of Union performance on PASCAL VOC 2012 (83.6% without MS-COCO pretraining), CamVid (76.4%) and Cityscapes (78.3%) datasets.

Recently, skeleton-based human action recognition has received more interest from industrial and research communities for many practical applications thanks to the popularity of depth sensors. A large number of conventional approaches, which have exploited handcrafted features with traditional classifiers, cannot learn high-level spatiotemporal features to precisely recognize complex human actions. In this paper, we introduce a novel encoding technique, namely Pose-Transition Feature to Image (PoT2I), to transform skeleton information to image-based representation for deep convolutional neural networks (CNNs). The spatial joint correlations and temporal pose dynamics of an action are exhaustively depicted by an encoded color image. For learning action models, we fine-tune end-to-end a pre-trained network to thoroughly capture multiple high-level features at multi-scale action representation. The proposed method is benchmarked on several challenging 3D action recognition datasets (e.g., UTKinect-Action3D, SBU-Kinect Interaction, and NTU RGB+D) with different parameter configurations for performance analysis. Outstanding experimental results with the highest accuracy of 90.33% on the most challenging NTU RGB+D dataset demonstrate that our action recognition method with PoT2I outperforms state-of-the-art approaches.

Despite impressive achievements in image processing and artificial intelligence in the past decade, understanding video-based action remains a challenge. However, the intensive development of 3D computer vision in recent years has brought more potential research opportunities in pose-based action detection and recognition. Thanks to the advantages of depth camera devices like the Microsoft Kinect sensor, we developed an effective approach to in-depth analysis of indoor actions using skeleton information, in which skeleton-based feature extraction and topic model-based learning are two major contributions. Geometric features, i.e. joint distance, joint angle, and joint-plane distance are calculated in the spatio-temporal dimension. These features are merged into two types, called pose and transition features, and then are provided to codebook construction to convert sparse features into visual words by k-means clustering. An efficient hierarchical model is developed to describe the full correlation of feature - poselet - action based on Pachinko Allocation Model. This model has the potential to uncover more hidden poselets, which have been recognized as the valuable information and help to differentiate pose-sharing actions. The experimental results on several well-known datasets, such as MSR Action 3D, MSR Daily Activity 3D, Florence 3D Action, UTKinect-Action 3D, and NTU RGB+D Action Recognition, demonstrate the high recognition accuracy of the proposed method. Our method outperforms state-of-the-art methods in the field in most dataset benchmarks.

In this paper, we propose a novel robust blind color image watermarking method, namely SMLE, that allows to embed a gray-scale image as watermark into a host color image in the wavelet domain. After decomposing the gray-scale watermark to component binary images in digits ordering from least significant bit (LSB) to most significant bit (MSB), the retrieved binary bits are then embedded into wavelet blocks of two optimal color channels by using an efficient quantization technique, where the wavelet coefficient difference in each block is quantized to either two pre-defined thresholds for corresponding 0-bits and 1-bits. To optimize the watermark imperceptibility, we equally split the coefficient modified quantity on two middle-frequency sub-bands instead of only one as in existing approaches. The improvement of embedding rule increases approximately 3 dB of watermarked image quality. An adequate trade-off between robustness and imperceptibility is controlled by a factor representing the embedding strength. As for extraction process, we exploit 2D Otsu algorithm for higher accuracy of watermark detection than that of 1D Otsu. Experimental results prove the robustness of our SMLE watermarking model against common image processing operations along with its efficient retention of the imperceptibility of the watermark in the host image. Compared to state-of-the-art methods, our approach outperforms in most of robustness tests at a same high payload capacity.

Novel interactive activity recognition method using a topic modeling technique.Pose-based spatio-temporal relation features for intra- and inter-person.Two types of codeword corresponding to joint distance and angle.Four-level Pachinko Topic Model (PAM) for flexibly modeling interactions.Outperformance of recognition accuracy to state-of-the-art methods. In this paper, we go beyond the problem of recognizing video-based human interactive activities. We propose a novel approach that permits to deeply understand complex person-person activities based on the knowledge coming from human pose analysis. The joint coordinates of interactive objects are first located by an efficient human pose estimation algorithm. The relation features consisting of the intra and inter-person features of joint distance and angle, are suggested to use for describing the relationships between body components of the individual persons and the interacting two participants in the spatio-temporal dimension. These features are then provided to the codebook construction process, in which two types of codeword are generated corresponding to distance and angle features. In order to explain the relationships between poses, a flexible hierarchical topic model constructed by four layers is proposed using the Pachinko Allocation Model. The model is able to represent the full correlation between the relation features of body components as codewords, the interactive poselets as subtopics, and the interactive activities as super topics. Discrimination of complex activities presenting similar postures is further obtained by the proposed model. We validate our interaction recognition method on two practical data sets, the BIT-Interaction data set and the UT-Interaction data set. The experimental results demonstrate that the proposed approach outperforms recent interaction recognition approaches in terms of recognition accuracy.