Search

Most facial analysis methods perform well in standardized testing but not in real-world testing. The main reason is that training models cannot easily learn various human features and background noise, especially for facial landmark detection and head pose estimation tasks with limited and noisy training datasets. To alleviate the gap between standardized and real-world testing, we propose a pseudo-labeling technique using a face recognition dataset consisting of various people and background noise. The use of our pseudo-labeled training dataset can help to overcome the lack of diversity among the people in the dataset. Our integrated framework is constructed using complementary multitask learning methods to extract robust features for each task. Furthermore, introducing pseudo-labeling and multitask learning improves the face recognition performance by enabling the learning of pose-invariant features. Our method achieves state-of-the-art (SOTA) or near-SOTA performance on the AFLW2000-3D and BIWI datasets for facial landmark detection and head pose estimation, with competitive face verification performance on the IJB-C test dataset for face recognition. We demonstrate this through a novel testing methodology that categorizes cases as soft, medium, and hard based on the pose values of IJB-C. The proposed method achieves stable performance even when the dataset lacks diverse face identifications.

Video-based person re-identification (ReID) aims to exploit relevant features from spatial and temporal knowledge. Widely used methods include the part- and attention-based approaches for suppressing irrelevant spatial-temporal features. However, it is still challenging to overcome inconsistencies across video frames due to occlusion and imperfect detection. These mismatches make temporal processing ineffective and create an imbalance of crucial spatial information. To address these problems, we propose the Spatiotemporal Multi-Granularity Aggregation (ST-MGA) method, which is specifically designed to accumulate relevant features with spatiotemporally consistent cues. The proposed framework consists of three main stages: extraction, which extracts spatiotemporally consistent partial information; augmentation, which augments the partial information with different granularity levels; and aggregation, which effectively aggregates the augmented spatiotemporal information. We first introduce the consistent part-attention (CPA) module, which extracts spatiotemporally consistent and well-aligned attentive parts. Sub-parts derived from CPA provide temporally consistent semantic information, solving misalignment problems in videos due to occlusion or inaccurate detection, and maximize the efficiency of aggregation through uniform partial information. To enhance the diversity of spatial and temporal cues, we introduce the Multi-Attention Part Augmentation (MA-PA) block, which incorporates fine parts at various granular levels, and the Long-/Short-term Temporal Augmentation (LS-TA) block, designed to capture both long- and short-term temporal relations. Using densely separated part cues, ST-MGA fully exploits and aggregates the spatiotemporal multi-granular patterns by comparing relations between parts and scales. In the experiments, the proposed ST-MGA renders state-of-the-art performance on several video-based ReID benchmarks (i.e., MARS, DukeMTMC-VideoReID, and LS-VID).

Multi-person pose estimation has been gaining considerable interest due to its use in several real-world applications, such as activity recognition, motion capture, and augmented reality. Although the improvement of the accuracy and speed of multi-person pose estimation techniques has been recently studied, limitations still exist in balancing these two aspects. In this paper, a novel knowledge distilled lightweight top-down pose network (KDLPN) is proposed that balances computational complexity and accuracy. For the first time in multi-person pose estimation, a network that reduces computational complexity by applying a "Pelee" structure and shuffles pixels in the dense upsampling convolution layer to reduce the number of channels is presented. Furthermore, to prevent performance degradation because of the reduced computational complexity, knowledge distillation is applied to establish the pose estimation network as a teacher network. The method performance is evaluated on the MSCOCO dataset. Experimental results demonstrate that our KDLPN network significantly reduces 95% of the parameters required by state-of-the-art methods with minimal performance degradation. Moreover, our method is compared with other pose estimation methods to substantiate the importance of computational complexity reduction and its effectiveness.

In recent times, as interest in stress control has increased, many studies on stress recognition have been conducted. Several studies have been based on physiological signals, but the disadvantage of this strategy is that it requires physiological-signal-acquisition devices. Another strategy employs facial-image-based stress-recognition methods, which do not require devices, but predominantly use handcrafted features. However, such features have low discriminating power. We propose a deep-learning-based stress-recognition method using facial images to address these challenges. Given that deep-learning methods require extensive data, we constructed a large-capacity image database for stress recognition. Furthermore, we used temporal attention, which assigns a high weight to frames that are highly related to stress, as well as spatial attention, which assigns a high weight to regions that are highly related to stress. By adding a network that inputs the facial landmark information closely related to stress, we supplemented the network that receives only facial images as the input. Experimental results on our newly constructed database indicated that the proposed method outperforms contemporary deep-learning-based recognition methods.

The decision tree is one of the most effective tools for deriving meaningful outcomes from image data acquired from the visual sensors. Owing to its reliability, superior generalization abilities, and easy implementation, the tree model has been widely used in various applications. However, in image classification problems, conventional tree methods use only a few sparse attributes as the splitting criterion. Consequently, they suffer from several drawbacks in terms of performance and environmental sensitivity. To overcome these limitations, this paper introduces a new tree induction algorithm that classifies images on the basis of local area learning. To train our predictive model, we extract a random local area within the image and use it as a feature for classification. In addition, the self-organizing map, which is a clustering technique, is used for node learning. We also adopt a random sampled optimization technique to search for the optimal node. Finally, each trained node stores the weights that represent the training data and class probabilities. Thus, a recursively trained tree classifies the data hierarchically based on the local similarity at each node. The proposed tree is a type of predictive model that offers benefits in terms of image's semantic energy conservation compared with conventional tree methods. Consequently, it exhibits improved performance under various conditions, such as noise and illumination changes. Moreover, the proposed algorithm can improve the generalization ability owing to its randomness. In addition, it can be easily applied to ensemble techniques. To evaluate the performance of the proposed algorithm, we perform quantitative and qualitative comparisons with various tree-based methods using four image datasets. The results show that our algorithm not only involves a lower classification error than the conventional methods but also exhibits stable performance even under unfavorable conditions such as noise and illumination changes., Competing Interests: The authors declare no conflict of interest.

Studies on depth images containing three-dimensional information have been performed for many practical applications. However, the depth images acquired from depth sensors have inherent problems, such as missing values and noisy boundaries. These problems significantly affect the performance of applications that use a depth image as their input. This paper describes a depth enhancement algorithm based on a combination of color and depth information. To fill depth holes and recover object shapes, asynchronous cellular automata with neighborhood distance maps are used. Image segmentation and a weighted linear combination of spatial filtering algorithms are applied to extract object regions and fill disocclusion in the object regions. Experimental results on both real-world and public datasets show that the proposed method enhances the quality of the depth image with low computational complexity, outperforming conventional methods on a number of metrics. Furthermore, to verify the performance of the proposed method, we present stereoscopic images generated by the enhanced depth image to illustrate the improvement in quality.

The research on hand gestures has attracted many image processing-related studies, as it intuitively conveys the intention of a human as it pertains to motional meaning. Various sensors have been used to exploit the advantages of different modalities for the extraction of important information conveyed by the hand gesture of a user. Although many works have focused on learning the benefits of thermal information from thermal cameras, most have focused on face recognition or human body detection, rather than hand gesture recognition. Additionally, the majority of the works that take advantage of multiple modalities (e.g., the combination of a thermal sensor and a visual sensor), usually adopting simple fusion approaches between the two modalities. As both thermal sensors and visual sensors have their own shortcomings and strengths, we propose a novel joint filter-based hand gesture recognition method to simultaneously exploit the strengths and compensate the shortcomings of each. Our study is motivated by the investigation of the mutual supplementation between thermal and visual information in low feature level for the consistent representation of a hand in the presence of varying lighting conditions. Accordingly, our proposed method leverages the thermal sensor's stability against luminance and the visual sensors textural detail, while complementing the low resolution and halo effect of thermal sensors and the weakness against illumination of visual sensors. A conventional region tracking method and a deep convolutional neural network have been leveraged to track the trajectory of a hand gesture and to recognize the hand gesture, respectively. Our experimental results show stability in recognizing a hand gesture against varying lighting conditions based on the contribution of the joint kernels of spatial adjacency and thermal range similarity., Competing Interests: The authors declare no conflict of interest.

In order to develop security systems for identity authentication, face recognition (FR) technology has been applied. One of the main problems of applying FR technology is that the systems are especially vulnerable to attacks with spoofing faces (e.g., 2D pictures). To defend from these attacks and to enhance the reliability of FR systems, many anti-spoofing approaches have been recently developed. In this paper, we propose a method for face liveness detection using the effect of defocus. From two images sequentially taken at different focuses, three features, focus, power histogram and gradient location and orientation histogram (GLOH), are extracted. Afterwards, we detect forged faces through the feature-level fusion approach. For reliable performance verification, we develop two databases with a handheld digital camera and a webcam. The proposed method achieves a 3.29% half total error rate (HTER) at a given depth of field (DoF) and can be extended to camera-equipped devices, like smartphones.

Vision-based hand gesture interactions are natural and intuitive when interacting with computers, since we naturally exploit gestures to communicate with other people. However, it is agreed that users suffer from discomfort and fatigue when using gesture-controlled interfaces, due to the lack of physical feedback. To solve the problem, we propose a novel complete solution of a hand gesture control system employing immersive tactile feedback to the user's hand. For this goal, we first developed a fast and accurate hand-tracking algorithm with a Kinect sensor using the proposed MLBP (modified local binary pattern) that can efficiently analyze 3D shapes in depth images. The superiority of our tracking method was verified in terms of tracking accuracy and speed by comparing with existing methods, Natural Interaction Technology for End-user (NITE), 3D Hand Tracker and CamShift. As the second step, a new tactile feedback technology with a piezoelectric actuator has been developed and integrated into the developed hand tracking algorithm, including the DTW (dynamic time warping) gesture recognition algorithm for a complete solution of an immersive gesture control system. The quantitative and qualitative evaluations of the integrated system were conducted with human subjects, and the results demonstrate that our gesture control with tactile feedback is a promising technology compared to a vision-based gesture control system that has typically no feedback for the user's gesture inputs. Our study provides researchers and designers with informative guidelines to develop more natural gesture control systems or immersive user interfaces with haptic feedback.

A light field camera is a sensor that can record the directions as well as the colors of incident rays. This camera is widely utilized from 3D reconstruction to face and iris recognition. In this paper, we suggest a novel approach for defending spoofing face attacks, like printed 2D facial photos (hereinafter 2D photos) and HD tablet images, using the light field camera. By viewing the raw light field photograph from a different standpoint, we extract two special features which cannot be obtained from the conventional camera. To verify the performance, we compose light field photograph databases and conduct experiments. Our proposed method achieves at least 94.78% accuracy or up to 99.36% accuracy under different types of spoofing attacks.

In this paper, we propose a new haptic-assisted virtual cane system operated by a simple finger pointing gesture. The system is developed by two stages: development of visual information delivery assistant (VIDA) with a stereo camera and adding a tactile feedback interface with dual actuators for guidance and distance feedbacks. In the first stage, user's pointing finger is automatically detected using color and disparity data from stereo images and then a 3D pointing direction of the finger is estimated with its geometric and textural features. Finally, any object within the estimated pointing trajectory in 3D space is detected and the distance is then estimated in real time. For the second stage, identifiable tactile signals are designed through a series of identification experiments, and an identifiable tactile feedback interface is developed and integrated into the VIDA system. Our approach differs in that navigation guidance is provided by a simple finger pointing gesture and tactile distance feedbacks are perfectly identifiable to the blind.

In this paper, a noble nonintrusive three-dimensional (3D) face modeling system for random-profile-based 3D face recognition is presented. Although recent two-dimensional (2D) face recognition systems can achieve a reliable recognition rate under certain conditions, their performance is limited by internal and external changes, such as illumination and pose variation. To address these issues, 3D face recognition, which uses 3D face data, has recently received much attention. However, the performance of 3D face recognition highly depends on the precision of acquired 3D face data, while also requiring more computational power and storage capacity than 2D face recognition systems. In this paper, we present a developed nonintrusive 3D face modeling system composed of a stereo vision system and an invisible near-infrared line laser, which can be directly applied to profile-based 3D face recognition. We further propose a novel random-profile-based 3D face recognition method that is memory-efficient and pose-invariant. The experimental results demonstrate that the reconstructed 3D face data consists of more than 50 k 3D point clouds and a reliable recognition rate against pose variation.

This paper presents a novel three-dimensional (3D) multi-spectrum sensor system, which combines a 3D depth sensor and multiple optical sensors for different wavelengths. Various image sensors, such as visible, infrared (IR) and 3D sensors, have been introduced into the commercial market. Since each sensor has its own advantages under various environmental conditions, the performance of an application depends highly on selecting the correct sensor or combination of sensors. In this paper, a sensor system, which we will refer to as a 3D multi-spectrum sensor system, which comprises three types of sensors, visible, thermal-IR and time-of-flight (ToF), is proposed. Since the proposed system integrates information from each sensor into one calibrated framework, the optimal sensor combination for an application can be easily selected, taking into account all combinations of sensors information. To demonstrate the effectiveness of the proposed system, a face recognition system with light and pose variation is designed. With the proposed sensor system, the optimal sensor combination, which provides new effectively fused features for a face recognition system, is obtained.

In this paper, we focus on the problem of the accuracy performance of 3D face modeling techniques using corresponding features in multiple views, which is quite sensitive to feature extraction errors. To solve the problem, we adopt a statistical model-based 3D face modeling approach in a mirror system consisting of two mirrors and a camera. The overall procedure of our 3D facial modeling method has two primary steps: 3D facial shape estimation using a multiple 3D face deformable model and texture mapping using seamless cloning that is a type of gradient-domain blending. To evaluate our method's performance, we generate 3D faces of 30 individuals and then carry out two tests: accuracy test and robustness test. Our method shows not only highly accurate 3D face shape results when compared with the ground truth, but also robustness to feature extraction errors. Moreover, 3D face rendering results intuitively show that our method is more robust to feature extraction errors than other 3D face modeling methods. An additional contribution of our method is that a wide range of face textures can be acquired by the mirror system. By using this texture map, we generate realistic 3D face for individuals at the end of the paper.