49 results about "Face hallucination" patented technology

Face hallucination using a bi-channel deep convolutional neural network (BCNN), which can adaptively fuse two channels of information. In one example, the BCNN is implemented to extract high level features from an input image. The extracted high level features are combined with low level details in the input image to produce the higher resolution image. Preferably, a proper coefficient is obtained to adaptively combine the high level features and the low level details.

Method, system and device for assisting diagnosis of autism which enable assistance of early detection and early definite diagnosis of autism (especially in infants) based on objective evaluation, using a conventionally suggested “eye-gaze detection technique”. The method for assisting diagnosis of autism of a subject uses an eye-gaze detecting unit, the unit having at least an imaging camera portion. The method includes displaying, at the position in front of the subject along the direction of the eye gaze of the subject, an illusionary image contained in a plane of an illusion-causing image on a displaying device. The eye-gaze position information of the subject looking at the plane is detected, and the eye-gaze position information of the subject is input into an eye-gaze position information storage portion The eye-gaze positions of the subject can be evaluated using a algorithm for assisting diagnosis of autism.

A method, system and computer readable media for real-time chromakey matting using image statistics. To identify the chroma key spectrum, the system / method executes in three stages. In an off-line training stage, the system performs semi-automatic calibration of the chroma key parameterization. In the real-time classification stage, the system estimates the alpha matte on a GPU. Finally, an optional error minimization stage improves the estimated matte, accounting for misclassifications and signal noise. Given the resulting matte, standard alpha blending composites the virtual scene with the video feed to create the illusion that both worlds coexist.

A data processing system provides high performance three-dimensional graphics. In one embodiment a processor of the data processing system performs a computer algorithm that creates a motion illusion regarding an object being displayed in a computer graphics scene by drawing multiple images of the object and varying the application of an attribute, such as transparency, color, intensity, reflectivity, fill, texture, size, and / or position including depth, to the images in a manner that provides the object with an illusion of three-dimensional motion between a first position and a second position. Also described are an integrated circuit for implementing the motion illusion algorithm and a computer-readable medium storing a data structure for implementing the motion illusion algorithm.

The invention discloses a human face super-resolution method based on residual keeping. According to the human face super-resolution method, based on a learning method, high and low resolution training image pairs are partitioned uniformly first, and processing is conducted in a block set based on positions: the smooth high frequency component and the standard deviation high frequency component are estimated on the corresponding positions of the training block set for all blocks of an input image, and the high-frequency compensation component is estimated through a residual training set. The linear combination of the three high-frequency information is used for reconstructing a high-resolution image on a high-resolution average face, and therefore the noise problem in super-resolution restoration of a human face image (like a monitoring image) with serious noisy points can be solved.

The invention discloses a face super-resolution method and device based on multi-view texture learning. The method belongs to the field of human face image super-resolution, and comprises the following steps: firstly, down-sampling a high-resolution human face image pair to a target low-resolution human face image pair, carrying out blocking operation on the target low-resolution human face imagepair, separating out mutually overlapped image blocks, and extracting facial texture multi-scale features by using a residual pooling module network; then, the extracted face multi-scale features aresent to a texture attention module, texture information is fused and compensated by calculating an attention map, the most similar features are collected, and the SR performance is more effectively improved. Finally, a feature map of the target view image is updated by feature fusion to produce a high resolution result. The network provided by the invention is superior to other latest face image super-resolution algorithms, and a face image with higher quality can be generated.

The invention discloses a human face super-resolution method and device based on a layered multi-scale residual fusion network, and belongs to the field of human face image super-resolution, and the method comprises the steps: downsampling a high-resolution human face image to a target low-resolution human face image, partitioning the target low-resolution image to acquire overlapped image blocksafter partitioning of the overlapped image blocks; extracting a fine facial feature map by using a bottleneck attention module; sending the extracted fine facial feature map to a multi-scale residualerror module, extracting feature information by using different convolution layers in the multi-scale residual error module to share feature information by using a cross mode, and achieve multi-scalefeature information fusion by using a jump connection mode outside the multi-scale residual error module so as to more effectively improve SR performance; and updating a feature map of the target low-resolution face image through feature fusion to generate a high-resolution result. The network provided by the invention is superior to other latest face image super-resolution algorithms, and a faceimage with higher quality can be generated.

The combination of a truly realistic three dimensional real image in color and a second device to produce a repetitive passage of a solid object through the real image creates the illusion of one solid body passing through another. A simple method of accomplishing this is to mount a clockworks with a pendulum over or under the image and allow a pendulum end to pass periodically through the real image. Other means such as a device running on a circular track could accomplish the same illusory effect. The pendulum end or the device can take various forms as can the real image. In principle the designs are not limited in size. Further, an amusement ride is contemplated by directing a track with a vehicle and a rider in the vehicle at the proper angle toward the real image creating an exciting visual moment passing through the apparently solid object.

A three-dimensional image simulation device for managing a live event comprising an image capturing device for capturing live captured data corresponding to a presenter and generating, in real-time, hologram data based on the live captured data. An output interface for broadcasting the hologram data in real-time to at least one additional location containing an audience, wherein the hologram data is used to create a hologram of the presenter at the at least one additional location based on an apparent parallax effect in a simulated three-dimensional display device, the hologram creating a three-dimensional illusion for the audience regarding actual presence of the presenter at the at least one additional location. Furthermore, an input interface for receiving audience data from the at least one additional location regarding interaction between the hologram and the audience and a display device for displaying images based on audience data to the presenter.

The invention relates to a local restriction iteration neighborhood embedding-based face hallucination method. The method comprises the following steps of: establishing high-resolution and low-resolution image block sets to be used as high-resolution and low-resolution image block dictionaries; sampling on inputted image blocks of a low-resolution face image to obtain estimation high-resolution image blocks, seeking K nearest image blocks at the corresponding position in the high-resolution image block dictionary, and expressing the inputted low-resolution image blocks by using the corresponding K low-resolution image blocks to acquire a weight coefficient; reconstructing K neighbor high-resolution image blocks by utilizing the weight coefficient to form new estimation high-resolution image blocks, and performing the operation repeatedly until the most satisfied estimation high-resolution image blocks are obtained; and integrating into a high-resolution image according to the positional relations of the low-resolution image blocks. According to the method, two manifold structures are considered simultaneously on the basis of position apriority and local manifold restriction, and K neighbor points and reconstruction weights are updated continuously in an iteration form on the basis of a result of last reconstruction to achieve a high-quality reconstruction effect which is close to the real condition.

The invention discloses a face super-resolution method and system based on a fusion attention mechanism, and belongs to the field of face image super-resolution. The method comprises the steps of down-sampling a high-resolution face image into a target low-resolution face image, carrying out partitioning operation, separating out image blocks which are overlapped with each other, and extracting superficial features by using a superficial feature extractor; fusing features of pixel, channel and space triple attention modules, and enhancing reconstructed face structure details; constructing a fusion attention network as a deep feature extractor, inputting the superficial facial features into the fusion attention network to obtain deep features, wherein the fusion attention network includes a plurality of fusion attention groups, each fusion attention group includes a plurality of fusion attention blocks; and performing up-sampling on the deep feature map, and reconstructing the face feature map subjected to up-sampling into a high-resolution face image of the target. The method is superior to other latest face image super-resolution algorithms, and face high-resolution images with higher quality can be generated.

A touch-based simulated false-vision positioning system in the technical field of visual prostheses comprises a helmet display, an auxiliary head positioning device, a touch screen, a computer, a touch navigation device and a touch panel. Firstly, the helmet display which is used as a stimulating source provides simulated false vision stimulation; with the auxiliary head fixing device, the head of a user is fixed opposite to where the touch screen is fixed; simulated false-vision generating software with an appropriate stimulating mode is selected and facilitates the computer to produce simulated false vision; after judging the positions of simulated false visions in the helmet display, the user points out the localization on the touch screen with the help of the touch navigation device or the touch panel; after feedback data is processed with corresponding software, analysis and evaluation on information such as localization accuracy, dispersion, etc. of simulated false visions sensed by touch can be obtained.

The invention relates to a face super-resolution method based on a supervised pixel-by-pixel generative adversarial network. The method comprises the following steps: firstly, reading an original facepicture data set; then carrying out preprocessing work such as data cutting and cleaning; thirdly, reading the high-resolution face image for bicubic interpolation down-sampling to obtain a high-resolution face image-low-resolution face image pair; fourthly, inputting the low-resolution face image into a generator network to generate a super-resolution face image; and fifthly, respectively inputting the high-resolution face image and the super-resolution image into a pixel-by-pixel discriminator network, calculating a supervised pixel-by-pixel adversarial loss function by using a pixel-by-pixel discrimination matrix output by the pixel-by-pixel discriminator network, and training a generator network by using error back propagation.

Identifying a masked suspect is one of the toughest challenges in biometrics that exist. This is an important problem faced in many law-enforcement applications on almost a daily basis. In such situations, investigators often only have access to the periocular region of a suspect's face and, unfortunately, conventional commercial matchers are unable to process these images in such a way that the suspect can be identified. Herein, a practical method to hallucinate a full frontal face given only a periocular region of a face is presented. This approach reconstructs the entire frontal face based on an image of an individual's periocular region. By using an approach based on a modified sparsifying dictionary learning algorithm, faces can be effectively reconstructed more accurately than with conventional methods. Further, various methods presented herein are open set, and thus can reconstruct faces even if the algorithms are not specifically trained using those faces.

This invention is related to displaying three-dimensional views and three-dimensional objects on a two-dimensional screen. More specifically, this invention is related to user interfaces of software displaying three-dimensional objects. The invention provides a way for the user to peek around his fingers, which are on top of the user interface or the augmented reality view. This can be implemented, for example, by using the front camera of the smart phone or the tablet to follow where the eyes and / or the face of the user is, and if the user moves his head the view displayed on the screen is changed to provide an illusion of a changed perspective so that the user would see what's beneath his fingers on the screen.

The invention discloses a face hallucination method and system based on a global-local separation attention mechanism. The method comprises the following steps: down-sampling a high-resolution face image to a target low-resolution face image; segmenting the target low-resolution image into blocks so as to extract a rough face feature map after mutually overlapped image blocks are obtained; constructing a separation attention network as a fine feature extractor, inputting the rough face features into the separation attention network to obtain a fine face feature map, wherein the separation attention network comprises a plurality of global-local separation attention groups, each global-local separation attention group generates two pieces of local attention and uses one global attention module to fuse different local attention, so that the local attention interacts across the feature groups, and the global attention of the network is realized; up-sampling the obtained face fine feature map; and reconstructing the up-sampled face feature map into a high-resolution face image of the target. According to the method, a face hallucination image with higher quality can be generated.

The invention discloses a robust target tracking method based on a phantom adversarial network, and relates to a computer vision technology. The method comprises the following steps: firstly, proposing a new illusion confrontation network, aiming at learning nonlinear deformation between sample pairs, and applying the learned deformation to a new target to generate a new target deformation sample.In order to effectively train the proposed phantom confrontation network, deformation reconstruction loss is proposed. The invention provides the target tracking method based on the illusion confrontation network based on the offline training, and the method can effectively alleviate the overfitting problem of the deep neural network in the target tracking process due to the online updating. In addition, in order to further improve the deformation migration quality, a selective variable migration method is provided, and the tracking precision is further improved. The target tracking method provided by the invention obtains a competitive result on the current mainstream target tracking data set.

Computer vision systems and methods for machine learning using image hallucinations are provided. The system generates image hallucinations that are subsequently used to train a deep neural network to match image patches. In this scenario, the synthesized changes serve in the learning of feature-embedding that captures how a patch of an image might look like from a different vantage point. In addition, a curricular learning framework is provided which is used to automatically train the neural network to progressively learn more invariant representations.

The present invention provides a face phantom method based on the nearest neighbor embedding of image position blocks. The high-rate face sample image is divided into overlapping image blocks, and the low-resolution and high-resolution face sample block spaces are established; for each image block in the input low-resolution face image, find the corresponding low-resolution image block dictionary K image blocks closest to the small image block, and then find the corresponding K adjacent high-resolution image blocks in the high-resolution image block dictionary for these K image blocks; use these K low-resolution image blocks to linearly represent Input a low-resolution image block to obtain a representation coefficient; use this representation coefficient and K adjacent high-resolution image blocks to reconstruct a new high-resolution image block; according to the positional relationship of the low-resolution image block, the resulting high-resolution image block into high-resolution images.

The invention discloses a face super-resolution method based on a dense residual attention face prior network. The method comprises the following steps: respectively constructing a dense residual attention module, a face structure prior prediction module and an up-sampling module which are connected by a jumper; connecting a dense residual attention module and a facial structure prior prediction module which are connected by a jumper in parallel, and then cascading an up-sampling module and an image reconstruction layer to construct a dense residual attention facial prior network; preprocessing the published data set, and dividing the processed data into a training set and a test set; training a dense residual attention face prior network; and inputting the images in the test set into thetrained dense residual attention face prior network, and outputting the reconstructed high-resolution face image. According to the method, the dense residual attention face prior network is trained, so that super-resolution processing of the tested face image is realized, high-frequency details of the face can be effectively recovered, and meanwhile, the super-resolution face image with identity information is reserved.

Method, system and device for assisting diagnosis of autism which enable assistance of early detection and early definite diagnosis of autism (especially in infants) based on objective evaluation, using a conventionally suggested “eye-gaze detection technique”. The method for assisting diagnosis of autism of a subject uses an eye-gaze detecting unit, the unit having at least an imaging camera portion. The method includes displaying, at the position in front of the subject along the direction of the eye gaze of the subject, an illusionary image contained in a plane of an illusion-causing image on a displaying device. The eye-gaze position information of the subject looking at the plane is detected, and the eye-gaze position information of the subject is input into an eye-gaze position information storage portion The eye-gaze positions of the subject can be evaluated using a algorithm for assisting diagnosis of autism.

The invention relates to low-resolution human face super-resolution, provides a super-resolution algorithm for solving the problem that expressions and postures of a reference image and a low-resolution human face image are greatly different, and discloses a human face super-resolution method guided by the reference image based on a three-dimensional deformation model. The method comprises the following steps of: firstly, deforming a reference image IR based on a three-dimensional deformation model; filling textures of an invisible area before deformation by utilizing the symmetry of a human face; eliminating artifacts after filling by using two operations at the same time; and taking the deformed reference image and the low-resolution image IL as the input of the super-resolution network,and outputting a high-resolution face image IH after passing through the super-resolution network. The invention is mainly applied to face recognition occasions.

The invention relates to the technical field of image recognition, in particular to a face image super-resolution calculation method based on deep learning. The method comprises the steps of 1, data processing; a reference image-based face super-resolution method is used for inputting data as a low-resolution image LR and a high-resolution reference image Ref, and an output target is a super-resolution image SR; 2, texture information is extracted; 3, texture information conversion; the texture information conversion module comprises a correlation coefficient matrix calculation module, a texture information query module and a texture weight assignment module; step 4, super-resolution processing based on the reference image; step 5, a loss function and an optimizer; step 6, evaluation method; step 7, verifying the effectiveness of the TFSR; and comparing the TFSR with the optimal face super-resolution method to verify the effectiveness of the TFSR.

The invention discloses a three-dimensional face super-resolution method based on multi-frame point cloud fusion and a variable model. The method comprises the following steps: S1, acquiring a video frame depth image and a point cloud sequence Pi belonging to P; s2, calculating a rough fitting result of the variable model by taking the average face as a template and Pi as a target point cloud, andobtaining a first rough fitting score; s3, screening the Pi according to the first rough fitting score to obtain a successfully detected point cloud set Pf; s4, taking a first point cloud P0 in the Pf as a target point cloud, taking all the rest face point clouds Pr in the Pf as templates, and respectively registering the P0 to obtain a second rough fitting score; s5, screening Pr according to the second rough fitting score, converting the screened point cloud to the position where P0 is located, and Palign = {P0, Pj0} is obtained; s6, converting the Palign to obtain a smooth fusion point cloud Pfusion; s7, performing variable fitting on the object Pk in the Palign by using a three-dimensional face variable model, and generating a variable model face fusion point cloud Mavg; and S8, fusing the Pfuse and the Mavg to obtain a three-dimensional human face super-resolution point cloud Poutput. According to the invention, the high-precision face point cloud can be obtained.

The invention discloses a video face hallucination method and a device based on component decomposition, which can improve the performance of the face hallucination in a low-resolution face video. The method comprises steps: as for each frame of image in the low-resolution face video for face hallucination, the face part in the frame of image is decomposed according to a face structure, and each face part obtained through decomposition is cut into image blocks; as for each image block obtained through cutting, a preset number of nearest neighbor image blocks for the image block is acquired from a preset image block library; as for each frame of image in the low-resolution face video, the nearest neighbor image blocks are used to adopt a local constraint-based neighborhood embedding method to rebuild the image block obtained through cutting the frame of image, high-resolution image blocks are obtained, and through combining the high-resolution image blocks, a target high-resolution image is obtained; and through combining each target high-resolution image, a target high-resolution face video is obtained.

The invention discloses a human face super-resolution method based on multi-scale attention residual errors and isovariant mapping. The method comprises the following steps: firstly, extracting shallow layer features of a low-resolution human face side image through a convolutional layer; then, inputting the shallow layer features into a feature extraction subnet, and obtaining deep layer features through n multi-scale attention residual modules; further inputting the obtained deep features into a residual isovariant mapping module, combining the deep features with residual features fused with yaw coefficients in a deep representation feature space, and transforming feature vectors of the side face into a feature vector space the same as that of the front face; and finally, a high-resolution face front image is obtained through a reconstruction module. The method is suitable for face recognition, does not excessively depend on a large number of front face and side face data pairs any more, and can reconstruct a front face image with high definition and more real texture for an input low-resolution side face image.

An optical illusion device and method of production is described. The method comprises producing a digital sculpt. A rendered visual image and a thinned digital sculpt are then produced from the digital sculpt. Next a digital thickness shell is produced from the thinned digital sculpt. Finally, the rendered visual image is employed to impart a lithophane depth map onto the digital thickness shell. The overall result is an optical illusion device comprising a negative image forming surface with a registered lithophane image imparted on thereon. When the optical illusion device is viewed from various angles (and with a light source placed behind it), the negative image forming surface appears to the viewer as a positive or convex model with a photographic image applied to the profiled surface. The illusion created is of a ‘panning’, three-dimensional photographic image, complete with improved detail due to the presence of the lithophane depth map.