26 results about "Artifact reduction" patented technology

The invention discloses an uncertainty perception simulation method based on a neural radiation field (NeRF), and aims to solve the problem of insufficient reality sense of a dynamic scene in an automatic driving scene. Aiming at the problems of rendering artifacts and uncertainty generated when a reflecting material and a shielding area are processed by a traditional simulation method, the invention provides a new framework fusing Bayesian uncertainty quantization and dynamic scene optimization. On the basis of an MARS simulator, the frame remarkably improves the visual quality and reliability of a simulation scene through a special rendering loss function and a physics-based rendering model. Experimental results show that the method is superior to a baseline MARS simulator in KITTI and VKITTI data sets, and particularly shows higher application value and generalization ability in the aspects of reducing artifacts and improving rendering fidelity, and provides a more reliable test environment for automatic driving simulation.

The invention discloses a sparse visual angle three-dimensional reconstruction method and system based on voxel grid constraint, and belongs to the technical field of visual three-dimensional reconstruction. Constructing a three-dimensional voxel grid of a self-adaptive scene scale based on point cloud distribution, and dividing the point cloud to the corresponding voxel grid; fusing the geometric features of the fast point feature histogram and the confidence score in the grid, generating a geometric confidence comprehensive measure, and screening key points to initialize Gaussian primitives; designing a voxel grid constrained gradient clipping strategy, limiting Gaussian primitive error diffusion through a distance attenuation coefficient, and adaptively optimizing grid distribution in combination with a dynamic grid deletion and addition mechanism; and finally, carrying out iterative training by using a 3D Gaussian splash radiation field loss function to realize high-fidelity static scene reconstruction under a sparse view angle. According to the method, scene geometric priori is introduced, an optimization strategy based on voxel grid constraint is designed to effectively control excessive diffusion or drift of Gaussian primitives, generation of artifacts is reduced, and meanwhile, the situation that robustness is reduced due to the influence of priori quality is avoided.

The present disclosure relates to systems, methods, and non-transitory computer-readable media that modify digital images via scene-based editing using image understanding facilitated by artificial intelligence. For instance, in one or more embodiments, the disclosed systems generate, utilizing a segmentation neural network and without user input, object masks for objects in a digital image. The disclosed systems determine foreground and background abutting an object mask. The disclosed systems generate an expanded object mask by expanding the object mask into the foreground abutting the object mask by a first amount and expanding the object mask into the background abutting the object mask by a second amount that differs from the first amount. The disclosed systems inpaint a hole corresponding to the expanded object mask utilizing an inpainting neural network.

The invention discloses an arbitrary style image migration method, computer equipment and a program product, the arbitrary style image migration method comprises the following steps: respectively extracting content features and style features from a content image and a style image, and fusing the content features and the style features to obtain input fusion features; inputting the style image into a feature extraction network comprising a normalization layer, respectively outputting a first style feature and a second style feature, performing feature fusion on the input fusion feature, the first style feature and the second style feature to obtain an output fusion feature, and using the output fusion feature to reconstruct a stylized image; and comparing the content feature loss of the stylized image and the content image, and the style feature loss of the stylized image and the style image, and training and optimizing the feature extraction network. According to the method, the content features and the style features in the spatial domain can be aligned in detail to further stylize, the texture coordination is improved, artifacts are reduced, and a high-quality stylized image is generated.

Disclosed are systems and techniques for image demosaicing with minimal artifacts. The techniques include computing a color value within a first color space for a first pixel of a plurality of pixels based at least on color values within the first color space of a first group of neighboring pixels of the first pixel and computing a first chrominance value within a second color space for the first pixel based at least on the computed color value within the first color space. The techniques include computing a luminance value within the second color space for the first pixel based at least on the first chrominance value within the second color space and converting the luminance value within the second color space and the first chrominance value within the second color space to an output pixel value within a third color space.

A pixel includes a semiconductor substrate that includes a floating diffusion region and a photodiode region. The pixel also includes, between a front surface of the semiconductor substrate and a back surface opposing the front surface: a first trench and a second trench adjacent to the first trench in a separation direction that is both (a) parallel to the front surface and (b) in a plane that is perpendicular to the front surface. Each of the first and second trench (a) is between the floating diffusion region and the photodiode region and (b) extends into the semiconductor substrate from the front surface. In the separation direction, a top average-separation between the first and second trench, at depths between the front surface and a first depth in the semiconductor substrate, exceeds a bottom average-separation between the first and second trench, at depths exceeding the first depth.

Disclosed are systems and circuits for image demosaicing with minimal artifacts. The circuits are to compute a color value within a first color space for a first pixel of a plurality of pixels based at least on color value within the first color space of a first group of neighboring pixels of the first pixel and compute a first chrominance value within a second color space for the first pixel based at least on the computed color value within the first color space. The circuits are also to compute a luminance value within the second color space for the first pixel based at least on the first chrominance value within the second color space and convert the luminance value within the second color space and the first chrominance value within the second color space to an output pixel value within a third color space.

Novel methods and systems for using conditional diffusion models to reduce video artifacts is realized. The method / system can utilize a pre-processed ground truth to improve the process, as well as multiple-frame conditioning, deformable convolutions, auxiliary enhancements, and adaptive temporal-spatial sampling techniques for improved performance. A quantization parameter agnostic framework is also possible.

The invention provides a CBCT metal artifact reduction method and system based on a coupling diffusion model, and belongs to the field of medical image processing. The method comprises the following steps: acquiring clinical data, and respectively training an MA CBCT image diffusion model and a clean CBCT image diffusion model based on the clinical data; training a noise conversion module by using the synthesized pairing data; inputting an MA CBCT image, and gradually adding noise to the MA CBCT image to an intermediate state through diffusion; mapping the noise features of the MA CBCT image to the noise space of the clean CBCT image diffusion model by using the trained noise conversion module; denoising step by step based on a noise level coupling mechanism; and generating an artifact reduction image through an MA adaptive reasoning fusion module. According to the method, full learning of metal artifact features is ensured, the problem that artifact feature conversion is inaccurate in a traditional method is solved, and the artifact removal effect is optimized.

A computer-implemented method of reducing artifacts in multi-channel magnetic resonance (MR) images is provided. The method includes receiving a plurality of sets of MR images acquired by a radio-frequency (RF) coil assembly having a plurality of channels. Each set of MR images includes a plurality of slices of MR images acquired by one of the plurality of channels. The method also includes estimating a plurality of sets of artifacts in the plurality of sets of MR images by inputting the plurality of sets of MR images into a neural network model. Each set of artifacts corresponds to the one of the plurality of channels. The method further includes reducing artifacts in the plurality of sets of MR images based on estimated artifacts, deriving MR images of reduced artifacts by combining the MR images of reduced artifacts, and outputting the MR images of reduced artifacts.

Nondestructive evaluation (NDE) of objects can elucidate impacts of various process parameters and qualification of the object. Computed tomography (CT) enables rapid NDE and characterization of objects. However, CT presents challenges because of artifacts produced by standard reconstruction algorithms. Beam-hardening artifacts especially complicate and adversely impact the process of detecting defects. By leveraging computer-aided design (CAD) models, CT simulations, and a deep-neutral network high-quality CT reconstructions that are affected by noise and beam-hardening can be simulated and used to improve reconstructions. The systems and methods of the present disclosure can significantly improve the reconstruction quality, thereby enabling better detection of defects compared with the state of the art.

The application discloses a kind of dual-domain constraint hyperspectral image reconstruction methods based on deep learning, it is related to the technical field of artificial intelligence and computational imaging.The application simultaneously imposes constraint in spectral reconstruction domain and sparse coefficient domain, while ensuring that its inherent sparse structure conforms to physical prior, the complement and verification of dual-domain information significantly improve the fidelity of reconstruction result, reduce artifact and noise;Using deep neural network to realize reconstruction, only once forward propagation is needed for new compression measurement value to output reconstruction result, which greatly reduces the computational complexity, improves the reconstruction speed, and has the potential for real-time processing;The sparsity physical prior is explicitly integrated into the network learning goal, which can effectively resist noise interference and reduce the influence of noise on the reconstruction result;Through back propagation algorithm, the dual-head deep neural network is optimized and trained end-to-end, and the best mapping relationship is automatically learned, which greatly reduces the operation difficulty and application threshold of the method.

An audio artifact reduction technique operates on a pair of temporally-adjacent audio tracks. Samples at an end portion of a first track may be altered according to an inter-track sample discontinuity at a boundary between the two tracks. Samples at a beginning portion of a second track also may be altered according to the inter-track discontinuity. These altering operations may cause the sample values and the end of the first track and the beginning of the second track to be equal to each other at the boundary between the two tracks. In one aspect, the altering may generate adjustment curves that alter sample values of the tracks around the boundary between them. In another aspect, the altering may trim the tracks at zero crossings of track sample values nearest to the boundary between the tracks.

This application discloses an image data processing method, apparatus, device, and storage medium, applicable to scenarios such as cloud gaming in cloud technology. The method includes: if a target basic block in a target image satisfies the block artifact reduction condition along with a first neighboring basic block in a first direction and a second neighboring basic block in a second direction, then a first boundary region and a second boundary region are determined within the target basic block; the target basic block, the first neighboring basic block, and the second neighboring basic block belong to different image blocks in the target image; single-linear interpolation is performed on the original pixel values ​​covered by the first boundary region to obtain a first image region; bilinear interpolation is performed on the original pixel values ​​covered by the second boundary region to obtain a second image region; and the target basic block after deblocking is determined based on the first image region and the second image region. Using this application embodiment can reduce block artifacts between basic blocks and improve image quality.

The application discloses a sequence image fusion method for rotation synthetic aperture computational imaging, and is based on the imaging mechanism of a rotation synthetic aperture optical remote sensing system, and proposes an end-to-end image fusion network based on a visual Transformer. Intra-frame self-attention calculation in a space-time information extraction module can more effectively process information of objects of different scales in a remote sensing image. At the same time, an inter-frame mutual attention is used to replace an explicit alignment module, so that the correlation between pixels at similar positions in different frames can be adaptively captured, and the generation of artifacts can be reduced. A visual sliding window Transformer module is used in a space-time information fusion module of the fusion network, time domain information is fully fused through the strong modeling capability of the Transformer itself, additional information in a low-quality image sequence can be fully utilized, and characteristics prior and data input are provided for actual on-orbit application of the rotation synthetic aperture system.

Systems and methods for artifact reduction in ultrasound imaging comprising acquiring a sequence of ultrasound images using an ultrasound probe along an acquisition plane, tracking a movement of the ultrasound probe along the acquisition plane and a movement of one or more artifacts in the sequence of ultrasound images, identifying one or more artifacts in the sequence of ultrasound images using the movement of the ultrasound probe along the acquisition plane and the movement of the one or more artifacts, wherein the movement of the one or more artifacts is different from the movement of the ultrasound probe along the acquisition plane, and correcting the sequence of ultrasound images, by at least one of adjusting a gain of the artifacts or stitching artifact areas of the sequence of ultrasound images with non-artifact areas of the sequence of ultrasound images.

Systems and methods for training a machine-learning model for artifact reduction are provided. Such methods include retrieving a three-dimensional digital phantom reconstructed from CT imaging data. The method then selects a first Z position along the central axis and simulates a first set of forward projections from the digital phantom taken along an axial trajectory at the first Z position along the central axis. The first set of forward projections has a first simulated collimation in the axial direction. The method then reconstructs a first simulated image from the first set of forward projections and identifies a plurality of secondary Z positions along the central axis other than the first Z position. For each of the secondary Z positions and the first Z position itself, the method then simulates a set of secondary forward projections from the digital phantom taken along corresponding axial trajectories at the corresponding secondary Z position.

A method includes acquiring a first volumetric image data set representing dentition of a patient, the first volumetric image data set including a modeled structure having an artifact distorting a boundary thereof, aligning a 3D digital impression of dentition of the patient with at least a portion of the first volumetric image data set, segmenting individual structures in the first volumetric image data set, merging the segmented individual structures to form a unitary volumetric model, thickening the 3D digital impression to form a shell bounding the at least a portion of the first volumetric image data set and supplementing the boundary of the at least one modeled structure, and, combining the 3D digital impression and the unitary volumetric model into a second volumetric image data set including the unitary volumetric model having at least a portion thereof bounded by the 3D digital impression.

This application relates to a spatial-frequency decoupled fluorescence-guided surgical image super-resolution reconstruction method, belonging to the field of image analysis technology. The method includes: constructing a super-resolution reconstruction network, which comprises a generator and a discriminator; the generator includes a shallow feature extraction module, multiple sequentially connected efficient two-stream decoupling structures, and a PixelShuffle upsampling layer; training the super-resolution reconstruction network based on a training dataset to obtain a trained generator; and inputting the low-resolution FGS image to be processed into the trained generator to obtain a super-resolution image. This application achieves synergistic optimization of FGS image noise suppression, artifact reduction, and detail preservation through two-stream decoupling and adaptive gating mechanisms, overcoming the drawback of generative adversarial networks (GANs) easily generating artifacts in medical images, resulting in reconstruction results that better meet realism requirements.

The invention relates to an image demosaicing technique with reduced artifacts, and specifically discloses a system and technique for image demosaicing with minimal artifacts. The techniques include calculating color values of a first pixel of a plurality of pixels within a first color space based at least on color values of a first set of adjacent pixels of the first pixel within the first color space; and calculating a first chromatic value of the first pixel in the second color space at least based on the calculated color value in the first color space. The techniques include calculating a luminance value of a first pixel within a second color space based at least on a first chromatic value within the second color space; and converting the luminance value within the second color space and the first chromatic value within the second color space into an output pixel value within a third color space.

For reconstruction in sampling-based imaging, such as reconstruction in MR imaging, an iterative, multiple-mapping based hierarchal machine-learned network reconstruction may produce artifact corrected images based on under-sampled scans. Two or more mappings may be used to reduce the presence of artifacts, in some cases including localized low-noise-contribution artifacts, relative to reconstructions based on fully-sampled scans.

The invention discloses a double-domain constraint hyperspectral image reconstruction method based on deep learning, and relates to the technical field of computational imaging and artificial intelligence. According to the method, constraints are applied to a spectrum reconstruction domain and a sparse coefficient domain at the same time, it is ensured that an internal sparse structure conforms to physical priori, the fidelity of a reconstruction result is remarkably improved through complementation and verification of double-domain information, and artifacts and noise are reduced; reconstruction is achieved through the deep neural network, a reconstruction result can be output only through one-time forward propagation for a new compression measurement value, the calculation complexity is greatly reduced, the reconstruction speed is increased, and the real-time processing potential is achieved. The sparse physical prior is explicitly integrated into a network learning target, noise interference can be effectively resisted, and the influence of noise on a reconstruction result is reduced; the end-to-end optimization training is performed on the double-end deep neural network through a back propagation algorithm, the optimal mapping relation is automatically learned, and the operation difficulty and the application threshold of the method are greatly reduced.

The edge prior and CNN-Transformer fused LDCT artifact suppression method of the present application belongs to the field of low-dose CT noise reduction. First, the transformer and convolution fusion module is introduced into the multi-scale coding and decoding network to extract texture and edge information of different scales of high-frequency parts. Second, in order to improve the edge perception of the high-frequency branch, the single-encoding double-decoding structure is used to construct the edge prior to supplement the multi-scale edge information for the high-frequency decoder. Finally, the high-frequency and low-frequency features of different scales at the decoding end are gradually fused and reconstructed into a denoised image through a multi-scale fusion module. In addition, multiple loss functions are designed to jointly constrain the training of the network to enhance the feature extraction capability of each branch and improve the stability of the network. The comparison and ablation test results on the Mayo LDCT dataset show the superiority of the method proposed in the present application in structure edge, texture preservation and artifact suppression.

The present application provides an ultrasonic time series data processing apparatus and a non-transitory storage medium. A Doppler processing section (18) or a beam data processing section (20) generates object time series data based on a received beam data string from a receiving section (16). An artifact prediction section (38) predicts a kind of artifact generated by the object time series data by inputting the object time series data to a learned artifact prediction learner (32). An artifact reduction section (40) performs artifact reduction processing based on the predicted kind of artifact. A display control section (24) notifies a user of a corrected ultrasonic image (62) on which the artifact reduction processing is performed or a prediction result of the artifact prediction section (38).

A computer-implemented method for artifact reduction in additive manufacturing includes monitoring an image of a printing process of an object being printed to detect stringing on the object and recognizing strings on the object in the image. The strings are mapped on the object on a map within a frame of reference relative to the object. The strings are removed from the object in accordance with the map.

The presently disclosed subject matter addresses the intrinsic limitation that artifact- free ground-truth X-ray images cannot be physically acquired, restricting the use of direct supervised learning for artifact correction. To overcome this constraint, the disclosure provides multiple mechanisms for developing machine-learning models dedicated to attenuating overlapping-layer artifacts in radiographs.