Recent deep learning based image editing methods have achieved promising results for removing object in an but fail to generate plausible large objects of complex nature, especially facial images. The objective this work is remove mask This problem challenging because (1) most the time masks cover quite a region face that even extends beyond actual boundary below chin, and (2) pairs with without do not exist training. We break into two stages: detection completion removed region. first stage...

The performance of face recognition methods using subspace projection is directly related to the characteristics their basis images, especially in cases local distortion or partial occlusion. In order for a method be robust and occlusion, images generated by should exhibit part-based representation. We propose an effective representation named locally salient ICA (LS-ICA) that LS-ICA only employs information from important facial parts maximize benefit applying idea "recognition parts". It...

This research presents a novel 2D feature space where real faces and masked fake can be effectively discriminated. We exploit the reflectance disparity based on albedo between materials. The vector used consists of radiance measurements forehead region under 850 685 nm illuminations. Facial skin mask material show linearly separable distributions in proposed. By simply applying Fisher's linear discriminant, we have achieved 97.78% accuracy face detection. Our method easily implemented...

Removing a specific object from an image and replacing the hole left behind with visually plausible backgrounds is very intriguing task. While recent deep learning based removal methods have shown promising results on this task for some structured scenes, none of them addressed problem in facial images. The objective work to remove microphone images fill correct semantics fine details. To make our solution practically useful, we present interactive method called MRGAN, where user roughly...

Obesity is a medical condition affecting billions of people. Various neuroimaging methods including magnetic resonance imaging (MRI) have been used to obtain information about obesity. We adopted multi-modal approach combining diffusion tensor (DTI) and resting state functional MRI (rs-fMRI) incorporate complementary thus better investigate the brains non-healthy weight subjects. The objective this study was explore use it predict practical clinical score, body mass index (BMI). Connectivity...

This research features a user-friendly method for face de-occlusion in facial images where the user has control of which object to remove. Our system removes one at time, however, it is capable removing multiple objects through repeated application. Although we show effectiveness our on five commonly occurring occluding including hands, medical mask, microphone, sunglasses, and eyeglasses, more types can be considered based proposed methodology. model learns detect user-selected, possibly...

Methods that enable the visual inspection of solar panels are currently in demand, as a huge number now being deployed sustainable energy source. One solutions for automation is an end-to-end deep learning framework, but this not recommended problem because such framework requires only powerful computational resources, also large-scale class-balanced dataset. In study, we present cost-effective panel defect detection method. We emphasize spatial feature defects by utilizing attention map...

Due to scarcity of anomaly situations in the early manufacturing stage, an unsupervised detection (UAD) approach is widely adopted which only uses normal samples for training. This based on assumption that trained UAD model will accurately reconstruct patterns but struggles with unseen anomalies. To enhance performance, reconstruction-by-inpainting methods have recently been investigated, especially masking strategy suspected defective regions. However, there are still issues overcome: (1)...

Process planning plays a key role by linking CAD and CAM. Its front-end is feature recognition, but recognition research has not been in accord with the requirements of process planning. This paper presents an effort for integrating two activities: feature-based machining sequence generation primarily based on tool capabilities. The system recognizes only manufacturable features consulting database, simultaneously constructs dependencies among features. Then, A* algorithm used to search...

Automatic monitoring of food intake in free living conditions is still an open problem to solve. This paper presents a novel necklace-type wearable system embedded with piezoelectric sensor monitor ingestive behavior by detecting skin motion from the lower trachea. Detected events are incorporated for classification. Unlike previous state-of-the-art based that employs spectrogram features, we have tried fully exploit time-domain signals optimal features. Through numerous evaluations on...

This research features a new idea for effective sinogram inpainting that boosts the MAR (metal artifacts reduction) performance. We have indentified neighbor pixels are relevant to target pixel be inpainted and only used these in determining an value. Experimental results show our methods based on proposed significantly reduce errors, enhancing They can incorporated into any sinogram-inpainting algorithms.

For defective solar panel detection, the use of resource-depleting methods such as end-to-end deep learning models does not serve purpose sustainable green energy. A recent study shows how this problem could be mitigated by exploiting attention-guided statistical features from an MNIST pre-trained attention map while achieving accurate defect detection panels. However, performance evaluation on mechanisms obtained different training datasets and neural network has never been reported. This...

Unsupervised anomaly detection (UAD) is a widely adopted approach in industry due to rare occurrences and data imbalance. A desirable characteristic of an UAD model contained generalization ability which excels the reconstruction seen normal patterns but struggles with unseen anomalies. Recent studies have pursued contain capability their models from different perspectives, such as design neural network (NN) structure training strategy. In contrast, we note that containing can also be...

Face recognition technology is widely used in law enforcement agencies. photo-sketch one of possible ways to identify suspects. We propose a method using joint dictionary learning for face recognition. Our bypasses the image synthesis procedure by previous based methods. Compared with other methods such as coupled which projects features from two different modalities into common space recognition, our does not need extra projections, and avoids expensive optimization learning. By cosine...

This research features a novel approach that efficiently detects depth edges in real world scenes. Depth play very important role many computer vision problems because they represent object contours. We strategically project structured light and exploit distortion of pattern the image along discontinuities to reliably detect edges. Distortion may not occur or be large enough depending on distance from camera projector. For practical application proposed approach, we have presented methods...

Image mosaicing is an image processing technique that most commonly used to conceal identities of sensitive objects. The authors’ research features recovering the mosaiced parts in image, especially focusing on facial parts. While recent completion methods based deep learning have shown promising results damaged they not addressed problem unmosaicing. Moreover, all those necessitate location information tackle recovery problem. They formulate unmosaicing as image‐to‐image translation...

This paper describes the application of a first order regularization technique to reconstruction visible surfaces. Our approach is computationally efficient method that simultaneously achieves approximate invariance and preservation discontinuities. It also robust with respect smoothing parameter /spl lambda/. The robustness property lambda/ allows free choice without struggling determine an optimal provides best reconstruction. A new approximately invariant stabilizing function for surface...

Image mosaicing conceals sensitive parts of an image. The objective this work is to recover hidden semantic structure under mosaiced parts, especially focusing on facial images. While recent image completion methods based deep learning have shown promising results recovering damaged in image, they not addressed the problem unmosaicing. We present a Generative Adversarial Network (GAN) approach unmosaicing called UMGAN, which image-to-image translation method. found that exploiting perceptual...

In the real world, different artists draw sketches of same person with artistic styles both in texture and shape. Our goal is to synthesize realistic face while retaining input identity, only using a single network. To achieve this, we employ modified conditional GAN target style label as input. method capable synthesizing multiple sketch even though it based on Sketches created by our show quality comparable state-of-the-art synthesis methods that use networks.

This research features a deep learning based framework to address the problem of matching given face sketch image against photo database. The photo-sketch is challenging because 1) modality gap between and very large, 2) number paired photo/ data insufficient train network. To circumvent large gap, our approach use an intermediate latent space two modalities. We effectively align distributions modalities in this by employing bidirectional (photo → photo) collaborative synthesis A...

In this paper, we present a novel collaborative bidirectional style transfer network based on generative adversarial (GAN) for cross modal facial image synthesis, possibly with large modality gap. We think that representation decomposed into content and can be effectively exploited synthesis. However, have observed unidirectional application of in case gap does not work well purpose. Unlike existing synthesis methods typically formulate as an feed forward mapping, our utilizes mutual...