27 results about "Focus stacking" patented technology

The invention relates to the technical field of image recognition, in particular to a construction site-oriented reinforcing steel bar binding normativity visual detection method, which comprises the following steps of: driving a camera lens to shoot a focus stack image sequence covering the depth of field before and after a single binding node, extracting a linear contour, and obtaining a multi-dimensional image set of a binding area. According to the method, the focus stack image sequence covering the depth of field before and after the single binding node is shot by the driving camera, and the mapping of the line segment coordinates and the depth of the reinforcing steel bar is established, so that the problems of mutual shielding and depth confusion caused by staggered reinforcing steel bars in the construction site are solved; on the basis, a steel bar binding topology network diagram describing the connection relation between steel bar intersection points is constructed, discrete steel bar line segments and intersection point information are integrated into an overall network with a global structure relation, detection is not limited to isolated nodes any more, and evaluation can be conducted from systematicness and continuity of a whole steel bar framework.

A method and device with depth map generation using focus stack data are provided. The electronic device includes one or more processors respectively comprising processing circuitry, and a memory storing code, which upon execution by the one or more processors, configures the one or more processors to generate focus stack data including images collected by an image collection device having a plurality of different focal lengths for a same scene at a plurality of viewing angles, generate a depth map for each of the images included in the generated focus stack data, by merging depth maps corresponding to an individual viewing angle among the generated depth maps, generate a single depth map for the individual viewing angle, and by processing and merging depth information of single depth maps generated corresponding to the plurality of viewing angles, generate a final depth map for the same scene.

The invention relates to the technical field of light field image salient target detection, in particular to a light field salient target detection method based on edge perception and hierarchical fusion, and the method comprises the steps: carrying out the multi-scale feature extraction of a focus stack image and a full-focus image through a backbone network; performing edge fusion enhancement on the focus stack features of four layers of different scales through an SEPM module and an EFM module; fusing high-level multi-modal semantic information from the global and local aspects by using an LHFM module; fusing low-layer space information and refining a salient target by using an LLFM module; and aggregating multi-scale information of a high layer and a low layer, and decoding the multi-scale information into an accurate saliency prediction image by using a detection head. According to the method, edge perception and a lightweight hierarchical fusion strategy are combined, the model parameter quantity and the calculation complexity are remarkably reduced while the high detection performance is kept, and the optimal balance between the performance and the efficiency is achieved.

The application provides a liquid lens focus stack imaging method, system and electronic equipment, and belongs to the technical field of optical imaging detection. The method comprises the following steps: a power supply module is controlled to provide a driving voltage for a liquid lens to form an image on a CMOS array; an image collected by the CMOS array is acquired, all images are aligned based on the image when the driving voltage is lowest, and an image sequence of a constant magnification focus stack is obtained; a focus state of the image sequence is evaluated, the image sequence is fused based on the focus state evaluation result, and a full-focus image is obtained; the object distance of the collected image is determined, a depth map is generated based on the object distance and the full-focus image, and a target image is generated based on the full-focus image and the depth map. The application aligns all collected images based on the image when the driving voltage is lowest, and solves the technical problem of low imaging accuracy existing in the existing focus stack imaging technology.

The invention relates to the technical field of image processing, and provides a large-depth-of-field real-time imaging method and system, and the method comprises the steps: controlling a liquid lens to adjust the focal length, shooting a plurality of initial images under each focal length through a camera, carrying out the alignment, classification and screening of all initial images, and carrying out the recognition of the initial images. Obtaining a target close-shot image, a target medium-shot image and a target long-shot image; and synthesizing the target close-shot image, the target medium-shot image and the target long-shot image by using the focus stack to obtain a target depth-of-field image. By controlling the curvature of the liquid lens, a plurality of images are quickly acquired on different focal lengths, each image is focused on different depth-of-field areas, a target close-shot image is synthesized by combining image alignment, classification, screening and a focus stack algorithm, frequent adjustment of the focal length of the lens and generation of a clear target depth-of-field image are not needed, and the image processing efficiency is improved. The problem that a traditional optical lens is low in efficiency when frequent focusing is needed is solved, and meanwhile the imaging range and definition are remarkably improved.

One among various embodiments discloses a computer-implemented method described above for generating an in-focus super-resolution image of a sample includes activating a light emitting element, wherein the light emitting element emits light towards a specimen in a growth medium, capturing a plurality of images of the specimen using an image sensor positioned on an opposing side of the specimen relative to the light emitting element, wherein each of the plurality of images is focused at a different depth, and generating an output image based on the plurality of images based on focus stacking the plurality of images to generate the output image.

The present invention discloses an image data parsing device, a scene estimation device, and a 3D fusion system, belonging to the field of video. It includes an image data parsing unit, an image raw data parsing unit, a focus stack data parsing unit, a camera parameter parsing unit, scene estimation, and image fusion for the remote fusion of multi-terminal real-scene data, realizing real-time, multi-terminal camera collaboration, and multi-application video image remote fusion, avoiding the problem of complex restrictions at the screen end. At the same time, without relying on a 3D rendering engine, it can reconstruct the real-scene data of multiple remote terminals, obtain the correct 3D visual geometric relationship, propose a solution for real-scene fusion, and avoid over-reliance on 3D virtual scene modeling operations. Moreover, it can freely change the viewing point while maintaining the consistent expression of the fused data. After the viewing point of the local terminal changes, the scene content of the fused data changes following the change of the local terminal's viewing point, thus achieving real-time seamless fusion between the remote and local scenes visually.

This invention discloses a distributed defocus stereo camera, relating to the fields of computational photography and depth measurement technology. Existing photogrammetry systems primarily obtain scene depth by acquiring sharp images or image stacks. Depth obtained using sharp images is often relative and lacks accuracy. Methods using focus stacks are complex, slow, costly, and impractical. This invention provides a low-cost, simple operation for acquiring dense and accurate depth, effectively solving this problem. The camera includes a distributed defocus acquisition device, a computational control system, and a depth measurement method. The acquisition device includes a beam splitter, a lens group, an aperture, a focusing device, a ranging device, and a photosensitive imaging device, used to acquire a sharp image, two blurred images, and the focusing distance of the scene. The computational control system calculates the scene depth using the scene data acquired by the distributed defocus acquisition device according to the corresponding depth measurement method.

This invention provides an adaptive image focus stacking method, system, and pathological slide scanner. It acquires multiple frames of original images from different focal planes within the same field of view, performs perceptual preprocessing on each frame, conducts multi-scale spatial frequency evaluation on each preprocessed frame, performs adaptive raster registration based on the multi-scale weight map, performs intelligent decision fusion on the corrected frames, and performs spatial consistency verification on the fused image, outputting a final full-area sharp focus stacked image. This invention offers advantages such as adaptive processing of high dynamic range images, improved robustness of sharpness evaluation, optimized alignment efficiency, and improved fusion effect, thereby generating high-quality full-area sharp focus stacked images.

Systems, devices and methods for sorting particles utilizing focus stacking of two-dimensional images are described. Such systems, devices and methods may further provide for particle processing and may encompass, on a microfluidic scale, sample enrichment, sample mixing, sample / particle sorting, verification of sorting and feedback-based optical sorting.

Implementations described herein relate to modifying depth of field in images using machine learning. In some implementations, a computer-implemented method for training a machine learning model includes generating an input training image that is a composition of multiple images captured in focus stacks at different lens focus positions and camera distances. A corresponding ground truth image is generated from merged images in particular focus stacks. A convolutional neural network (CNN) machine learning (ML) model receives the input training image and outputs an output image that adjusts blurriness in the input training image to simulate a target depth of field. The CNN ML model is updated based on comparison of the output image and the ground truth image. The CNN ML model can include a depth CNN that performs an implicit depth estimation for features of the input image, and a deconvolution CNN that adjusts the blurriness.

The invention discloses a metal product surface defect detection method and system based on light field imaging. The detection method comprises the following steps: acquiring a light field image; based on the light field image, focus stack image features and full-focus image features are acquired; based on the focus stack image feature and the full-focus image feature, obtaining a residual feature outside a focal plane; acquiring defect features based on the residual features outside the focal plane; fusing the defect features and the full-focus image features to obtain comprehensive features; based on the comprehensive features, a defect prediction map is obtained, and the defect prediction map is used for defect positioning and detection. According to the method, the focusing stack and the full-focusing image are effectively processed, so that high-precision detection on the surface defects of the metal product is realized.

The invention relates to the technical field of fundus detection, in particular to a fundus three-dimensional topography measurement system and method based on human eye dynamic response coding, which can realize novel fundus three-dimensional topography measurement which does not depend on a precise mechanical structure, does not need binocular registration, and is simple to assemble and operate, simple in system and controllable in cost. The dynamic adjustment capability of a crystalline lens of a human eye is used as an imaging driving mechanism, the human eye is guided to continuously adjust the focal length by projecting a dynamic virtual image with far and near changes, a focus stack is generated through coding for three-dimensional reconstruction, finally, a fundus color picture and three-dimensional structure data of the fundus color picture are synchronously obtained, and accurate measurement of the fundus three-dimensional shape is achieved. The method does not need complex registration operation of binocular stereo imaging and image texture dependence during three-dimensional reconstruction, and also does not need complex scanning mechanisms and precision machines of laser scanning ophthalmoscopes, optical coherence tomography and other technologies, so that the system hardware cost and the operation difficulty are reduced, and the method is easier to popularize and apply in basic medical scenes.

Implementations described herein relate to modifying depth of field in images using machine learning. In some implementations, a computer-implemented method for training a machine learning model includes generating an input training image that is a composition of multiple images captured in focus stacks at different lens focus positions and camera distances. A corresponding ground truth image is generated from merged images in particular focus stacks. A convolutional neural network (CNN) machine learning (ML) model receives the input training image and outputs an output image that adjusts blurriness in the input training image to simulate a target depth of field. The CNN ML model is updated based on comparison of the output image and the ground truth image. The CNN ML model can include a depth CNN that performs an implicit depth estimation for features of the input image, and a deconvolution CNN that adjusts the blurriness.

The application discloses a kind of based on multi-process focus stack image fusion method and system, belong to industrial blemish detection and photography field, including the following steps: quickly acquire the mth, m=1,2,......,N, N is the total number of image, the image of the same object is shot;In the application, traditional image edge detection method is changed into multi-image gradient edge detection method based on multi-process, so as to realize the synthesis of focus stack image, which greatly improves the running time of synthesis, especially when CPU is in multi-core condition, while the method also has the processing process for image, which can effectively improve the quality of acquired image, thereby effectively improving the subsequent focus stacking effect, ensuring the quality of final focus stack image, and in the end, image evaluation process is also provided, which can learn high-quality images according to deep learning algorithm, compare and evaluate, and evaluate the overall focus stack image fusion method, which is powerful and suitable for promotion.

The invention provides a flame three-dimensional multi-physics field reconstruction method based on a light field focus stack, which comprises the following steps of: imaging blackbody radiation sources at different temperatures by using a light field camera, and establishing and outputting a fitting relationship between response values of color channels of a light field refocusing image and corresponding spectral radiation intensities according to a Planck law; acquiring a flame light field focus stack image sequence of the flame light field original image by adopting a super-resolution reconstruction method based on a sub-aperture image; and outputting temperature distribution and soot concentration distribution of the corresponding flame slice image by using the three-dimensional convolutional neural network model constrained by the fused radiation transmission physical information. According to the method, the implicit correlation model between the flame focus slice and the three-dimensional physical field of the flame focus slice is directly constructed by fusing the three-dimensional convolutional neural network model constrained by the forward radiation transmission physical information, and the generalization ability and the interpretability of the model can be improved, so that the joint reconstruction work of the physical field is more effectively realized.

The invention relates to an image acquisition system, in particular to an image acquisition device and an image acquisition method for multi-focus stacked macro photography. The invention discloses an image acquisition device for multi-focus stacked macro photography. The image acquisition device comprises a mobile device; the shooting seat is arranged on the mobile device and is used for fixing a shot object; the shooting equipment is arranged opposite to the shot object, and the shooting equipment and the moving direction of the shot object are located on the same axis; and the controller is respectively connected with the moving device and the shooting equipment, and is used for controlling the moving device to drive the shot object to move from the shooting starting point to the shooting ending site, and controlling the shooting equipment to shoot a picture of the current position when the moving device moves for each set length. The object to be shot is placed on the movable shooting seat, and in view of the fact that the object to be shot is generally light in weight, the mobile device and the control system of the mobile device can realize miniaturization and light weight, so that the whole set of device is convenient to operate and move, simple and convenient to use and not easy to shake.

The present invention discloses a YOLOv8 dense pedestrian detection method based on a novel focused stacking block FocalNeXtFold, including the following steps: Step S1: Obtain a dataset with the training set, test set, and validation set partitioned; Step S2: Improve the YOLOv8 network by replacing all C2f modules in the Backbone backbone network with focused stacking blocks FocalNeXtFold Block; Step S3: Based on Step S2, replace the head part with a network that can strengthen the direct interaction between non-adjacent layers; Step S4: Based on Step S3, replace the loss function of the YOLOv8 network to complete the improvement of the model; Step S5: Use the partitioned dataset to train the improved YOLOv8 model in Step S4 to obtain a trained model; Step S6: Use the trained model to detect the data to be detected and output the detection result. By replacing the backbone C2f block with the novel focused stacking block FocalNeXtFold, and forming the FocalNeXtFold stacking block by stacking several FocalNeXt block blocks, the dense pedestrian detection performance is improved.

Methods and apparatus apply focus stacking to sample preparation, improving accuracy of analytic tasks, facilitating automation, and improving throughput. Focus stacking is applied to a set of sample images having different focus depths, to produce a composite image in which features at different depths are in focus and, optionally, a depth map. A sample location is selected from the composite image and a localized material removal, measurement, or imaging operation is performed based on the sample location. A depth value from the depth map is used to set a working depth of a tool for performing the localized operation. Applications include lamella preparation for cryogenic TEM analysis of biological samples. Other applications, techniques, and variations are disclosed.

The present application relates to the technical field of image acquisition, in particular to a super-DOF image acquisition system and method for defect detection, which comprises a light source module, a lens, a prism, a camera and a displacement driving assembly, the displacement driving assembly is fixedly connected with a vertical mounting plate, the light source module, the lens, the prism and the camera are sequentially docked from bottom to top, and are fixedly installed with the surface of the vertical mounting plate as an installation reference surface, the optical axes of the light source module, the lens, the prism and the camera are collinear, the light source module is composed of LED lamps with different heights, different colors and different angles, and the camera and the light source module are connected with an external host computer. The method comprises focus stacking, image definition evaluation, image alignment and fusion optimization. The present application can overcome the limitations of fixed-focus cameras, obtain clear full-view images of devices with height differences, better meet the image acquisition requirements in defect detection, and further promote the progress of defect detection technology.

A method for 3D imaging of a sample using a machine learning algorithm is disclosed. The method uses a multi-modal focus stack consisting of a plurality of images acquired at two or more distances between the sample and a front focal plane, where at least one image is acquired using a first modality and an additional image is acquired using an additional modality. The modalities may have different illumination angles and optionally have different spectral distributions. The method may include receiving a training image of a sample including a plurality of training focus stacks and live 3D data (depth map) for each training focus stack; and training a machine learning algorithm based on the training image and the live data. The method then receives a product image of the sample, the product image including a stack of focuses from the optical assembly; and generating a 3D depth map.

The application discloses a focus depth estimation method and device based on an event focus stack, which comprises the following steps: taking a full-focus image and a corresponding scene depth map of a target scene by using a camera, calculating a defocus blur quantity, and post-rendering the full-focus image to obtain a focus stack, and simulating the focus stack to obtain an event focus stack; preprocessing the event focus stack by using a voxel grid and a depth surface coding method to obtain an event tensor, and constructing a multi-scale neural network to evaluate the focus degree of the event focus stack and estimate the scene depth, wherein the multi-scale neural network comprises a focus information extraction module, a depth regression module and a depth enhancement module; using the estimated scene depth to train the multi-scale neural network in a supervised framework to obtain a target multi-scale neural network; and inputting the event tensor into the target network model to reconstruct a scene depth image to estimate the target scene depth.

The invention discloses an ultrasonic hardened layer detection method and system based on wave velocity inversion and migration imaging, and the method comprises the steps: S1, carrying out the full-matrix data collection of a sample through employing an ultrasonic phased array technology, and obtaining reflection signals under different transmitting-receiving channel combinations; s2, constructing a common midpoint set model based on the reflected signal, determining a hyperbolic trajectory based on the common midpoint set model, and generating a speed-energy distribution map based on the hyperbolic trajectory; s3, based on the speed-energy distribution atlas, using a migration imaging method of a Kirchhoff delay superposition principle to obtain a speed-energy distribution atlas after delay compensation domain focusing superposition; s4, ultrasonic hardened layer detection based on wave velocity inversion and migration imaging is completed on the basis of the speed-energy distribution atlas after focusing and superposition of the delay compensation domain.

An example method includes determining that a portion of a scene in an image frame being captured at a first focal length is out of focus. The method also includes capturing one or more first image frames at the first focal length and one or more additional image frames at a second focal length to focus on the portion of the scene. The method additionally includes providing the one or more first image frames and the one or more additional image frames as input to a machine learning (ML) model, the ML model having been trained to merge one or more focused regions in a plurality of input images to predict an output image with an extended depth of field (DoF). The method also includes receiving the predicted image from the ML model.

The application provides a high-fidelity hologram generation method based on a focus stack network, which comprises six steps: in the first step, a full-focus image and a depth image of a 3D object are processed into a focus stack by using a focus stack renderer; in the second step, a focus stack network is used to output a complex amplitude distribution of a hologram of the focus stack of the 3D object; in the third step, a holographic reconstruction image is generated by using an angular spectrum diffraction propagation model; in the fourth step, a target in-focus and out-of-focus image and a reconstructed in-focus and out-of-focus image are calculated; in the fifth step, a loss function is calculated, and a focus stack network and a learnable Zernike phase are optimized; the above five steps are repeated, and when the focus stack network is iteratively trained for a fixed number of rounds, the training of the focus stack network is stopped, and the output of the second step is the required complex amplitude distribution; in the sixth step, the complex amplitude distribution of the focus stack is compensated by using the Zernike phase, and then a high-fidelity hologram of the 3D object is generated by using a two-phase encoding method.

The invention discloses a hologram model training method and device based on an unlabeled sample training set, equipment and a medium, and relates to the field of three-dimensional imaging, and the method comprises the steps: obtaining an unlabeled sample training set containing a plurality of RGB-D images; performing iterative training on the hologram model by using the unlabeled sample training set until a preset number of iterations is reached; in the training process, for each RGB-D image in the training set, calculating a multi-depth diffraction field matrix of the RGB-D image under a preset diffraction distance, and inputting the multi-depth diffraction field matrix into a hologram model to generate a multi-depth hologram; performing layer-by-layer diffraction reconstruction on the multi-depth hologram by using an angular spectrum method to obtain a focal stack image; and comparing the difference between the target RGB-D image and the focus stack image to calculate the model loss, and optimizing the parameters of the hologram model according to the model loss. According to the method, the hologram model is trained through the label-free sample training set, the dependence on the data set is reduced, and the coding quality and the generation speed of the three-dimensional multi-depth hologram are improved.

The application discloses a metal product surface defect detection method and system based on light field imaging. The detection method comprises the following steps: acquiring a light field image; acquiring a focus stack image feature and a full-focus image feature based on the light field image; acquiring a residual feature outside a focal plane based on the focus stack image feature and the full-focus image feature; acquiring a defect feature based on the residual feature outside the focal plane; fusing the defect feature and the full-focus image feature to acquire a comprehensive feature; and acquiring a defect prediction map based on the comprehensive feature, wherein the defect prediction map is used for defect positioning and detection. Through effective processing of the focus stack and the full-focus image, high-precision detection of the metal product surface defect is realized.