102 results about "Image registration algorithm" patented technology

The invention discloses a thermal imaging temperature measurement and fault location inspection system and relates to the field of automatic detection. The inspection system comprises a video server platform and a client platform which are connected with each other, wherein the video server platform is used for acquiring and preprocessing an invisible light image and a thermal infrared image of an inspection field and sending the preprocessed invisible light image and thermal infrared image to the client platform; and the client platform is used for processing the received invisible light image and the thermal infrared image by using an image registration algorithm and a thermal imaging temperature measurement algorithm and displaying the processed invisible light image and thermal infrared image, and sending a control command to the video server platform. By the image registration and thermal imaging temperature measurement method, the precision of fault location and temperature measurement is improved.

The invention discloses an image super-resolution rebuilding method based on a variable focal length video sequence. The image super-resolution rebuilding method includes the following steps: a first step is that a group of low-resolution images with different focal lengths are shot to form one video sequence, all the images are changed into gray level images and subjected to image preprocessing, and reference images are selected from the gray level images; a second step is that matching point pairs between the reference images and the remaining unselected images are acquired by using the scale invariant feature transformation algorithm; a third step is that a homography matrix between the reference images and the remaining unselected images is calculated according to the matching point pairs by using the random sampling consensus algorithm; and a fourth step is that the reference images are subjected to super-resolution rebuilding by using the maximum posterior probability algorithm. Through the sub pixel precision image registration algorithm, the image super-resolution rebuilding method allows that translation, rotation, zooming and other situations exist among the images, has an outstanding effect in the super-resolution rebuilding of the variable focal length video sequence, and has certain innovativeness.

The invention is applicable to the field of image processing and provides an estimation method and system for a lung motion model. The method includes: extracting a control point in a lung three-dimensional CT image of a reference phase; using a tracking position at a previous moment of the control point as a prediction position and tracking the position of the control point in a continuous-discrete-phase three-dimensional CT image; performing least-square curve fitting on the continuous-discrete-phase pursuit path of the control point and determining the motion position of the control point at any continuous time; and constructing a three-dimensional deformation model of the lung at a current time according to the motion position of the control point at any continuous time and obtaining lung motion position estimation of any voxel of the lung of the reference phase at any time. The estimation method and system for the lung motion model adopts an image registration algorithm based on a mark point so that the method is small in calculation quantity, high in calculation speed and capable of estimating the motion model of the lung effectively.

The invention discloses a fusion method of a multi-robot three-dimensional geometrical map, comprising the steps of projecting a three-dimensional geometrical map to be fused into two-dimensional grid maps; fusing the grid maps by adopting an image matching algorithm to acquire rotation parameters among the two-dimensional grid maps; and rotating and transforming the three-dimensional map and fusing the three-dimensional geometrical map by adopting a three-dimensional point set registration algorithm. The fusion method of the multi-robot three-dimensional geometrical map integrates the two-dimensional map fusion and the three-dimensional point set registration algorithm through considering geometrical characteristic information in the three-dimensional map to finish the fusion of the multi-robot three-dimensional geometrical map and obviously improve the fusion accuracy of the three-dimensional geometrical map. The fusion method of the multi-robot three-dimensional geometrical map is mainly applied to simultaneously positioning a plurality of robots and reestablishing the map and fusing partial map created by a single robot to form a global map.

The invention relates to a zero point calibration method and the application in an optical micro scanning micro thermal imaging system thereof, belonging to the infrared thermal imaging field. The method is as follows: firstly, two images are collected by taking the testing zero point of a rotating platform as a starting point; and then the micrometric displacement between the two images is calculated by using an improved frequency domain image registration algorithm; the rotating angle and direction is determined by a micro zero point calibration method according to the micrometric displacement; the rotating platform is controlled to rotate a corresponding angle to reach the zero point position of optical micro scanning of the system. And then four under sampling and low resolving power images are collected according to a standard 2*2 micro scanning mode and over-sampling images, the resolving power of which is increased by one time, are obtained by directly interpolation according to a mode of collecting images. As a micro scanner device of a monoblock infrared parallel plate, hardware with high positioning accuracy, fast speed and being apt to the system, and a real time implementation processed micro scanning zero point calibration method are adopted, the resolving power of the space of the system is obviously increased, thus promoting the application of micro thermal imaging technique in multiple fields.

The invention relates to the technical field of image processing, in particular to a remote sensing image registration algorithm. The registration algorithm comprises the following steps of: (1) reading primary remote sensing image data from a storage medium; (2) pre-processing each primary image; (3) performing binarization processing to realize image segmentation; (4) corroding and expanding the binary image; (5) extracting a monotropic edge from the eroded and expanded binary image by the Roberts cross gradient operator; (6) calculating the matching degree of the edge image and finding the optimal matching point; (7) conversing images; and (8) outputting registration information and registration images. The remote sensing image registration algorithm can be applied to multispectral data fusion and remote sensing image mosaicking.

The invention discloses a 3D (three-dimensional) image registration method and device based on conformal geometric algebra. The 3D image registration method comprises the following steps of: rebuilding a position relation restrain problem of a 3D medical image by using the conformal geometric algebra, analyzing conformal geometric transformation of the medical image, constructing a new 3D medical image registration similarity measure, and proposing a 3D medical image registration algorithm for 3D registration of CT (Computed Tomography) and MR_T1 (Magnetic Resonance) images. According to the three-dimensional image registration method and the three-dimensional image registration device, the direct alignment of 3D data is realized, the 3D positions of tissues and organs can be better positioned, the registration result is more direct, the image after registration is clearer, and the registration precision is higher.

The invention discloses multi-source image registration based on a local structure binary pattern and belongs to the technical field of image processing. The multi-source image registration based on the local structure binary pattern includes the following steps that first, feature points are detected and positioned; second, the feature points are filtered out through saliency; third, the main direction is determined; fourth, descriptors are established based on local structure binary pattern operators; fifth, the feather points are matched. Compared with a classic image registration scale invariant feature transform (SIFT) algorithm which is poor in effect on images from different sources, the same threshold is adopted in the matching step, multiple matches can be generated and are all correct, and thus the multi-source image registration has high registration accuracy.

The invention relates to an image fusion system based on a low-light and infrared sensor. The system is divided into a low-light and infrared sensor information acquisition module, an FPGA core processing module, a DSP co-processing module and an output display module; video images collected by the external low-light and infrared sensor are input to a device through an interface; the input video images are subjected to image preprocessing including image denoising, image enhancement and image registration, wherein image denoising and image enhancement are low in delay requirement and are achieved in an FPGA chip, the computation amount required for an image registration algorithm is large, and image registration is achieved in a DSP chip; the pre-processed video images are fused through the FPGA chip, wherein grayscale fusion and pseudo-color fusion are included; an FPGA control logic can select one path of images to be output through the interface, wherein one path of images include asingle low-light video, a single infrared video, a grayscale fused video and a pseudo-color fused video. The image fusion system is high in precision, low in delay and good in expansibility and has important practical significance on nighttime observation and exploration application.

This algorithm provides a marker-free approach to establishing the pixel correspondence among the IR images taken at different times, which is the basis for quantitatively characterizing the variation of the heat energy and patterns pixel-wise on a breast surface. The idea is to use the corner points of the heat pattern and the branch points of the skeletons of the heat pattern on the body surface as the initial fiducial points for the longitudinal IR image registration. The Thin-Plate Spline technique is used to model the nonlinear deformation between two IR images taken at two different times. Mutual information between the TPS-transformed image and the target image is employed as the metric quantifying the quality of the longitudinal IR image registration. To optimize the registration, Nelder-Mead simplex method is used to locally modify the pairings of the fiducial points in the source and target IR images to maximize the mutual information.

The invention discloses a medical image registration algorithm based on a convolutional neural network. The algorithm specifically comprises the following steps: data acquisition and preprocessing; Making a neural network training data set; Training a neural network; Performing coarse registration on the neural network model; And fine registration is carried out based on a gray value registrationalgorithm. According to the invention, the convolutional neural network can be skillfully combined with the traditional gray value-based registration algorithm; The invention discloses a convolutionalneural network elimination method in medical image processing. the complete labeled input sample is lacked; Training sample limitation and other problems, the spatial difference between a group of input fixed images and floating images is quickly calculated; Meanwhile, a neural network model rapidly obtains the spatial difference between the input images to be registered, the large spatial difference is reduced, the situation that a local optimal solution can be obtained in the optimization iteration process based on a gray value registration algorithm is avoided, and therefore the spatial difference between the two images to be registered can be rapidly and accurately calculated.

The image registration systems and methods according to this invention register reflectance and fluorescence hyperspectral imagery and provide the spatial image transformation between the reflectance and fluorescence hyperspectral imagery. A reflectance image is embedded into the fluorescence hyperspectral imagery and thereby resolves the resemblance problem of reflectance and fluorescence hyperspectral imagery. The reflectance image is embedded into the fluorescence hyperspectral imagery using (1) light from a narrow-band source embedded in Ultraviolet (UV) excitation light, (2) second order diffracted light, and / or (3) Ultraviolet (UV) excitation light reflectance. Resembling reflectance images are calculated from the reflectance hyperspectral imagery and the embedded reflectance image in the fluorescence hyperspectral imagery. The two respective reflectance images are pre-processed to maximize the resemblance between them. Image registration algorithms use the two pre-processed reflectance images and calculate the spatial image transformation that allows to map the spatial coordinates from the reflectance to the fluorescence hyperspectral imagery (and vice versa).

The invention discloses a Bayes' rule based multi-frame blind convolution super-resolution reconstruction method and device. The method includes steps of acquiring an interest area of a reference image and a matching area of a target image subjected to radiation correction through an image quality evaluation and frame selection algorithm for an image sequence of one scene; acquiring the radiancy and an accurate geometric distortion parameter through an image registration algorithm executed on the matching area of the target area subjected to radiation correction; acquiring a point diffusion function of image super-resolution restoration through execution of a multi-frame blind deconvolution image restoration algorithm on the accurate geometric distortion parameter; acquiring a super-resolution reconstruction image through execution of a maximum posterior super-resolution reconstruction algorithm on the radiancy and the point diffusion function of image super-resolution restoration. According to the invention, problems of insufficient consideration of point diffusion function, motion blur, image structure information, sparsity and the like of a traditional general algorithm are solved, automatic estimation on the system point diffusion function and multi-frame image registration parameters is performed, and the image resolution is improved.

The invention brings forward a novel image registering method. The method comprises the following steps: establishing gray-scale-based mutual information registering adaptation value function; establishing and initializing a group, wherein four dimensions of each individual in the group respectively represents horizontal translation, vertical translation, a rotation angle and a zoom coefficient of a floating image; according to the mutual information registering adaptation value function, calculating a fitness value of each individual, and calculating an optimal position of the whole group; by use of an iteration mechanism of a differential evolution algorithm, updating a position vector of each individual, and updating the optimal position of the whole group; determining whether conditions for executing an alternative strategy is satisfied, and if so, executing the corresponding alternative strategy; and repeatedly executing the aforementioned steps until maximum iteration frequency Tmax of the differential evolution algorithm is satisfied. The novel image registering method has the advantages of good registering stability and high precision, greatly improves the performance of an image registering algorithm and lays a reliable foundation for subsequent image processing work.

The invention relates to an image registration method based on SIFT (scale invariant feature transform) feature and angle relative distance. The technical feature is characterized by, to begin with, extracting feature points and feature point descriptors in a reference image and an image to be registered through an SIFT feature point extraction algorithm; then, obtaining a coarse matching result between image feature points through Euclidean distance between the feature points; and with the coarse matching result being input, calculating angle relation of the feature points, and carrying out subtraction to obtain angle difference and screening out correctly-matched feature point pairs. Correctly-matched feature point screening can be carried out by fully utilizing corresponding angle invariant feature of the coarsely-matched feature point pair set, thereby greatly improving correctly-matched feature point pair screening capability, and improving accuracy of image registration.

The invention discloses a non-rigid image registration algorithm based on implicit shape representation and edge information fusion. The method comprises the following steps: firstly, the external contours of a floating image and a reference image to be registered are aligned to realize the global registration based on implicit shape; secondly, a free-form deformation model of multi-resolution mesh of B spline based on edge information fusion is selected, and local registration is performed. The algorithm has the advantages that by introducing the measuring to edge registration in a registration framework, the important contribution of edge information in image registration is fully utilized, the registration accuracy and robustness are improved and smooth and continuous transform domains ensuring one-to-one mapping are obtained.

The invention pertains to the technical field of screen detection, and provides a screen detection system and method. The system comprises: an electromigration platform for setting mark feature points on a detected target at a certain spacing distance; an image acquiring device for performing shooting on the detected target at a certain interval time and make the mark feature points exist in the shot adjacent images; and an image processing device for performing the image splicing fusion process through the mark feature points in the shot images and acquiring a screen detection result according to the processed image. According to the screen detection system and method of the embodiments of the invention, by setting specific marks on the detected target as panoramic splicing key points before acquiring the images, the number of the feature points can be reduced, the efficiency of the image registration algorithm can be improved, and according to the invention, multiple images can be spliced at the same time, and the splicing process can be performed by utilizing multiple cameras. Through the image result after the splicing fusion process, a screen is determined whether to be a standard screen, so the screen detection technology is more simple and effective, and the industrial detection cost can be greatly reduced.

The invention discloses a multi-frame super-resolution image reconstruction method and system. An advanced image registration algorithm is adopted to enhance the image registration accuracy and application is more flexible and convenient, and a denoising link is additionally arranged after image reconstruction so that noise amplified in the super-resolution process can be removed and the sharpness of super-resolution images can be enhanced.

The invention relates to the technical field of medical auxiliary recognition and relates to a focus recognition method for ischemic stroke magnetic resonance images. The method comprises an image input step, a DWI image processing step, an ASL image processing step, an infusion calculation step, a low perfusion calculation step, a DWI three-dimensional reconstruction step, an ASL three-dimensional reconstruction step, and an ischemia semi-dark zone identification step and the like. In the DWI image processing step and the ASL image processing step, a morphological method is adopted to enhance the characteristics of the image. Focus segmentation and direct volume calculation are realized based on a segmentation method of an adaptive threshold value. Ischemia semi-dark band recognition is realized through an image registration algorithm, and then a Mismatch value is calculated. The cerebral arterial thrombosis lesion position can be accurately recognized, a specific lesion volume value is provided, three-dimensional visual data display is provided, an accurate reference basis is provided for diagnosis and treatment of doctors, and clinical decision making of the doctors is assisted.

The invention discloses a three-channel CMOS synchronous polarization imaging system, which belongs to the technical field of photoelectric imaging detection. The system comprises a polarization optical system, a three-channel CMOS image acquisition module, a DSP image processor module and an image registration algorithm. The polarization optical system is composed of linear polarization films, electric three-variable lenses and a driving module thereof. The installation directions of linear polarization of three acquisition channels are respectively the 0-degree direction, the 45-degree direction and the 90-degree direction. The three-channel CMOS image acquisition module is composed of three CMOS image sensors, completes image acquisition under the maximum resolution and can be applicable to static image capture and continuous video acquisition. The DSP image processor module can simultaneously process a plurality of channels of digital video streams. The image registration algorithm includes feature point detection, feature point matching, transformation model estimation and image transformation and reacquisition. The three-channel CMOS synchronous polarization imaging system is capable of better detecting water surface tracer particles and provides an effective particle detection effect for large-scale particle image velocity measurement.

The invention relates to a medical image registration algorithm evaluation method based on a deformation field. According to the method, the deformation field generated in the registration process acts on template split images to obtain deformation split images, and finally evaluation parameters of all regions are obtained. The method specifically includes the steps that (1) a template image and a fixed image are selected, and registration is carried out by applying a registration algorithm to be evaluated; (2) the deformation field generated in the registration process is extracted and acts on one template split image to generate a deformation split image, and the template split image is split into N parts; (3) the deformation split image and a fixed split image are utilized to calculate regional evaluation parameters RSCR; (4) the fixed image is changed, the first step, the second step and the third step are repeated to obtain 39 sets of regional evaluation parameters, a regional evaluation parameter set and an overall evaluation parameter are figured out and led out of a box plot, and ultimately the evaluation result is obtained. Compared with the prior art, the method has the advantage that the registration effect of a certain region can be evaluated through the rectification algorithm.

The invention provides an infrared and visible light information fusion navigation method based on bionic vision. The method at least comprises the following steps: (1) image enhancement preprocessing: processing a visible light gray scale gradient problem and a noise problem of an infrared image in an acquired image by adopting a Gauss-Laplace operator; (2) image registration, wherein the main purpose of image registration is to remove or suppress inconsistency of geometric space between an image to be registered and a reference image; (3) image fusion: simulating an ON confrontation system by using a Gaussian function to obtain an enhanced image, simulating the confrontation system by using a logarithmic function to obtain a suppressed signal image, then forming a pseudo-color image, andperforming color distortion correction; and (4) visual navigation. The navigation method has the excellent characteristics of strong anti-interference performance, good imaging capability and high resolution, and meanwhile, the image registration algorithm based on the salient region is adopted, so that the registration real-time performance is high, and the navigation algorithm is high in speed.

The invention relates to the field of medical image post treatment, for the purpose of providing a rapid convex optimization algorithm based method for registering three-dimensional CT and ultrasonic liver images. The rapid convex optimization algorithm based method for registering the three-dimensional CT and ultrasonic liver image images comprises the following process: adjusting the resolution of ultrasonic images and CT images to be the same; performing rigid transformation based coarse registration on the ultrasonic images and the CT images; extracting unified characteristic information of multimodal image registration; calculating a D(u) in a data item and a partial D(u) of the D(u) under the condition of a non-rigid deformation field u(x); performing model solving on each step of a gradual convex optimization method to obtain a deformation field optimal rectification value h(x), and updating the deformation field until the h(x) is very small; and according to the solved non-rigid deformation field, transforming the ultrasonic images for registration with the CT images. According to the invention, through establishment of a reasonable model, a rapid and accurate three-dimensional ultrasonic-CT liver image registration algorithm is designed, and the accuracy, the safety and the effectiveness of an ablation operation are improved.

The invention discloses a motion correction method for magnetic resonance multiple excitation diffusion imaging. The method comprises the steps of collecting a magnetic resonance multiple excitation sequence; obtaining an image echo signal and a two-dimensional navigation signal through a magnetic resonance scanning sequence; estimating motion parameters through the two-dimensional navigation signal; performing rotational and translational correction on the image echo signal and the two-dimensional navigation signal according to the motion parameters, and discarding tainted data, thereby obtaining corrected data; integrating the corrected data collected in multiple excitation to perform parallel imaging reconstruction; estimating rotational motion parameters through an image registration algorithm to perform diffusion gradient correction on a reconstructed image; and performing final calculation by utilizing corrected diffusion image and diffusion gradient to obtain diffusion tensor imaging parameters. According to the method, various motion errors in a magnetic resonance diffusion imaging scanning process can be effectively corrected, so that an artifact-free high-resolution diffusion tensor image is obtained, the errors are effectively reduced, and the accuracy of calculating magnetic resonance diffusion imaging tensor parameters is improved.

The invention discloses a brain magnetic resonance image registration method based on a K-means clustering method. The method comprises the steps that firstly, an accumulated angle point set is acquired through mixed angle point detection of a Harris operator and a SUSAN operator, a new angle point set is screened out through angle point strength, rough matching screening is carried out through a cross correlation coefficient method, then angle points are clustered by introducing the K-means clustering method, precise angle point pairs are screened out by combining a normalized correlation method and a voting matching method, finally, the angle point set is optimized through a Powell algorithm, a rebuilt parameter value is obtained, and final registration is carried out on images. The brain magnetic resonance image registration method has the advantages that good stability is achieved, feature point information can be described completely and accurately, the running time of a program in subsequent operation is reduced, registration accuracy is improved, an image registration algorithm is improved more precisely and efficiently, and a transformation image fluctuant within a large range can be compatible better.

The invention relates to a multi-temporal POLSAR image registration method based on SAR-SIFT features. In the multi-temporal POLSAR image registration process, a multi-dimensional polarization scattering vector is constructed by combining polarization scattering components to calculate the gradient of a pixel point. According to the method, the problem that polarization information in the image cannot be fully utilized during POLSAR image registration in the prior art is solved. The feature point extraction precision in the registration process is effectively improved, a registration algorithmhas high performance, and an efficient registration task is achieved. In a process of constructing a scale space during multi-temporal POLSAR image registration, the total variation model is adoptedto replace Gaussian convolution, so that more detail information is reserved in the constructed scale space, speckle noise in the image can be effectively filtered out, stable feature points are extracted, mismatching point pairs are eliminated, and the accuracy and robustness of the multi-temporal POLSAR image registration algorithm are improved.

The invention discloses a steel mesh image registration and detection method for a large-format optical measurement system, relates to the field of image processing, and solves the problems of low registration and detection precision and high difficulty of an existing image registration and detection method. According to the method, an image registration algorithm is guided by taking a prior design parameter of the large-format optical measurement system as prior information, creatively a star map identification method is introduced into a steel mesh image registration method, and three key parameters including a zooming parameter, a rotation parameter and a translation parameter required to be calculated for steel mesh image registration are separately calculated with different methods, so that a complicated steel mesh image registration problem is simplified. According to the method, the design parameter of the large-format optical measurement system is fully utilized as the prior information of image detection, so that the difficulty in originally complicated steel mesh image registration and detection is lowered; and the star map identification method is introduced into the steel mesh image registration method for the first time and mutually verified with a registration method using Hough transformation and a prior placement angle, so that the success rate of registration is effectively ensured.

The invention relates to an imager capable of integrating full color and polarization hyperspectral detectability, comprising a front lens, a field diaphragm, a collimating lens, an acousto-optic tunable filter and a turning lens which are successively arranged on the same light path; after light is turned by the turning lens, a polarization hyperspectral imagining system is arranged on the light path of level-one diffraction light; and light is turned by the turning lens, the light path of level-zero diffraction light is provided with a full-color light imaging system. The invention provides a programmable polarization hyperspectral and full-color combined imager which splices an O-light image with an E-light image on the photosensitive surface of the same detector by an optical splicing device, causes the system structure to be compact by light path turning, carries out image drift compensation by an optical wedge compensating device and an image registration algorithm, realizes multi-frequency acousto-optic tunable filter (AOFT) driving based on any waveform generator scheme and can simultaneously obtain full-color image information and narrow-band cross polarization image information.

The invention discloses a dominant measurement and implicit measurement integrating multi-mode image registering method. The dominant measurement and implicit measurement integrating multi-mode image registering method is aimed to solve the problem of easiness in mismatching due to large initial error of a characteristic mapping-based image registering algorithm. The dominant measurement and implicit measurement integrating multi-mode image registering method includes that using mutual information as dominant measurement, using the particle swarm optimization to obtain initial image conversion parameters, and using the characteristic mapping-based registering algorithm to further optimize the parameters to obtain more precise image registering parameters after converting the image to be registered. In order to fasten the image registering speed and improve the registering precision, an image pyramid structure is used for performing layered optimization to solve. The dominant measurement and implicit measurement integrating multi-mode image registering method is suitable for multi-mode image registering, is capable of improving the image registering success rate and enables the image registering error to be reduced.