51results about How to "Rich texture details" patented technology

The invention discloses an image denoising method based on a generative adversarial network, and the method is characterized by comprising the following steps: 1, selecting an experiment data set; 2,selecting Gaussian additive white noise as a noise model; 3, building a generative network model, and training a generator network G for denoising; 4, establishing a discrimination network model, wherein a discriminator D is used for carrying out authenticity classification on the input image; 5, constructing a joint loss function model; 6, training a generative adversarial network; and step 7, carrying out image denoising quality evaluation. According to the image denoising method based on the generative adversarial network, the denoising effect of reserving more texture details and edge features can be achieved.

The invention provides an attention mechanism-based image blind deblurring method and system, and the method comprises the steps that a multi-scale attention network directly recovers a clear image inan end-to-end mode, wherein the multi-scale attention network adopts an asymmetric encoding and decoding structure, and the encoding side of the encoding and decoding structure adopts a residual dense network block to complete feature extraction and expression of an input image by the multi-scale attention network; a plurality of attention modules are arranged on the decoding side of the encodingand decoding structure, the attention modules output preliminary restored images, and the preliminary restored images form a pyramid-type multi-scale structure of the image; the attention module further outputs an attention feature map, and the attention feature map models the relationship between the long-distance areas from the global perspective to process the blurred image; dark channel priorloss and multi-scale content loss form a loss function, and the loss function is used for reversely optimizing the network and is not self-optimized.

The invention discloses a face image correction method based on a decoupling expression learning generative adversarial network. The method comprises the following steps: a model comprising an auto-encoder of a U-net network structure and three discriminators is trained; then, identity information representation of the face image is learned through the auto-encoder, the posture of the face image can be generated through explicit control in combination with the posture hidden code, and the three discriminators predict authenticity, posture and identity information of the face image respectively, so that the face image with rich texture details is generated. According to the invention, the visual quality of the generated face image can be significantly improved.

The invention relates to an NSST domain flotation froth image enhancement and denoising method based on a quantum harmony search fuzzy set. The NSST domain flotation froth image enhancement and denoising method comprises the steps: carrying out NSST decomposition on a flotation froth image, and obtaining a low-frequency sub-band image and multi-scale high-frequency sub-bands; performing quantum harmony search fuzzy set enhancement on the low-frequency sub-band image; secondly, for the multi-scale high-frequency sub-bands, removing a noise coefficient by combining an improved BayesShrink thresholding and scale correlation, and enhancing an edge coefficient and a texture coefficient through a nonlinear gain function; and finally, performing NSST reconstruction on coefficients of the processed low-frequency sub-band and each high-frequency sub-band to obtain an enhanced de-noised image. According to the NSST domain flotation froth image enhancement and denoising method, the brightness, the contrast and the definition of the froth image can be improved, and the froth edge is obviously enhanced while noise is effectively inhibited, and more texture details are reserved, and subsequent processing such as froth segmentation and edge detection is facilitated.

The present invention relates to an adaptive image inpainting method based on a continued fraction interpolation technology and a system thereof. Compared to the prior art, the defects of bad inpainting effect and low efficiency are solved. The method comprises the following steps: initialing image characteristic analysis; employing a unitary continued fraction interpolation technology to perform inpainting of damaged points of a scratched image; employing a binary continued fraction interpolation technology to perform inpainting again of the scratched image, taking an initial inpainting image A as an information image, determining the position of each damaged point through a mask artwork, employing the binary continued fraction interpolation to update pixel information around the damaged points to obtain pixel information of each damaged point, and obtaining a final inpainting image B. The adaptive image inpainting method based on the continued fraction interpolation technology and the system thereof improve the quality and efficiency of image inpainting, can perform inpainting of any types of scratches, and are suitable for processing of all the images.

The invention relates to an image processing method and device, electronic equipment and a computer readable storage medium, and the method comprises the steps: obtaining a first image, coding the first image to extract features, and obtaining a corresponding coding feature map; and obtaining random noise, and performing image reconstruction on the coded feature map based on the random noise in a decoding process to obtain a second image, wherein the second resolution of the second image is greater than the first resolution of the first image. By adopting the method, the texture details of the image can be increased in the super-resolution reconstruction process of the image, so that the texture details of the image are richer.

The invention discloses an infrared video enhancement system based on multi-level guided filtering, and the system comprises an FPGA module and a DSP module. The FPGA module and the DSP module are incommunication connection through an SRIO interface, wherein the FPGA module completes digital single green video data acquisition and transmits the data to the DSP module through the SRIO interface; after video enhancement processing is completed in the DSP module, the video data is transmitted back to the FPGA module through the SRIO interface, and finally the video data is output to an externalscreen display through an LVDS interface of the FPGA module after being converted into a format through the FPGA module. A framework system of the DSP module and the FPGA module is adopted, and the single-frame image processing is efficient. The video transmission is achieved between the DSP module and the FPGA module through an SRIO interface, and the system has the advantages of being high in bandwidth, high in efficiency, high in real-time performance, low in time delay and the like. According to the present invention, the multi-level guided filtering is achieved through box filtering, thecalculation complexity is irrelevant to the size of a window, clear structural characteristics and rich texture details can be reserved, and the system is suitable for occasions with high requirementsfor real-time performance.

The invention provides an accurate and small-computation remote sensing image super-resolution reconstruction method combining a two-parameter beta process with a dictionary, and belongs to the technical field of remote sensing image processing. The method comprises the following steps: S1, inputting a low-resolution image to be reconstructed, a high-resolution image dictionary D(x), a low-resolution image dictionary D(y) and a mapping matrix A; S2, performing sparse coding on the input low-resolution image according to a dictionary D (y) to obtain a low-resolution sparse coefficient, mappinga high-resolution sparse coefficient corresponding to the low-resolution sparse coefficient by using a matrix A, and reconstructing a high-resolution image by using the high-resolution sparse coefficient and a dictionary D (x); the acquisition method of the dictionary D (x), the dictionary D (y) and the matrix A comprises the steps that according to paired high-resolution and low-resolution training images, the dictionary D (x), the dictionary D (y) and the matrix A are acquired through a double-parameter beta process, the matrix A is a mapping matrix from sparse coefficients of the dictionaryD (y) to sparse coefficients of the dictionary D (x), and the sparse coefficients are products of coefficient weights and dictionary atoms.

The invention relates to a damaged bank note image positive and negative repair method and a damaged bank note image positive and negative repair system based on a continued fraction interpolation technology. Compared with the prior art, the defect that the image repair technology lacks of stability and universality and cannot be used to repair image details is solved. The method comprises the following steps: making an initial image feature analysis; repairing the damaged points of a note image using a positive repair way; repairing the damaged points of the note image using a negative repair way; and constructing a final repair result C using the two repaired images. The quality and efficiency of image repair are improved.

The invention discloses a latent fingerprint label-free Stokes parameter polarized confocal microimaging system and method. The system comprises a polarization generation module, a confocal scanning module, a polarization measurement module and a data collection imaging module. According to the system, fingerprint polarization scanning imaging is carried out based on Stokes parameter measurement,and Stokes parameter and polarized parameter fingerprint images can be obtained after a monochromatic plane polarized light passes through an object printed with fingerprint marks; the background brightness formed by transmission of an objected, such as a mirror, can be well reduced, so that the fingerprint has remarkable contrast with the background, the image background interference can be weakened, and the fingerprint differentiation degree can be improved; and the fingerprint image has rich texture details and shows clear featured details of primary feature, secondary feature and partial tertiary feature which is hard to observe in other fingerprint imaging methods. The method is simple and easy to operate, does not need processing on a fingerprint sample, and can avoid destroying fingerprints.

The invention relates to a three-dimensional dress-up model processing method and device, computer equipment and a storage medium. The method comprises the steps of reducing a surface quantity of an initial three-dimensional dress-up model; mapping a three-dimensional dress-up model subjected to the surface quantity reduction into a two-dimensional dress-up texture grid in a texture coordinate system; adding texture information to the two-dimensional dress-up texture grid, thereby obtaining a corresponding two-dimensional dress-up texture image; projecting the two-dimensional dress-up textureimage to the surface of the three-dimensional dress-up model subjected to the surface quantity reduction; generating a normal map according to the difference between the initial three-dimensional dress-up model and the three-dimensional dress-up model subjected to the surface quantity reduction; and on the surface of the projected two-dimensional dress-up texture image of the three-dimensional dress-up model subjected to the surface quantity reduction, attaching the normal map to obtain a corresponding three-dimensional dress-up model. Through the projected two-dimensional dress-up texture image added with the texture information, and in combination with the normal map of texture details of the model with a high surface quantity, the final three-dimensional dress-up model has a lot of richtexture details, and the efficiency is improved.

The invention provides an image processing method and device, a computer readable storage medium and electronic equipment, and relates to the technical field of image processing. The image processing method comprises the following steps: acquiring a first image and a second image shot for the same scene; wherein the first image is shot by a first camera, the second image is shot by a second camera, and the light incoming amount of the first camera is greater than that of the second camera; performing region segmentation on the second image by using the first image to obtain at least one first object region of the second image; and performing noise reduction on the first object area to obtain a second object area corresponding to the first object area. The image quality can be improved.

The invention relates to a no-reference low-illumination image enhancement method based on a deep convolutional neural network, and belongs to the field of image processing. The method comprises the following steps: firstly, constructing a feature extraction module comprising two branches by using a deep convolutional neural network, and extracting an irradiation component and a reflection component from an input low-illumination image; and then, denoising the reflection component, and integrating the denoised reflection component into an optimization network to obtain an optimized reflection component; then, inputting the irradiation component into the optimization network to obtain an optimized irradiation component; and finally, multiplying the optimized irradiation component by the reflection component to obtain a final enhancement result. According to the method, the reflection component extracted from the input image is fully utilized, the interference of noise in the image is effectively reduced, and the detail expression capability is improved.