72 results about "Color moments" patented technology

The invention discloses a method for extracting a landmark scene abstract, which comprises the following steps of: extracting color moment and ripple lamination of each landmark scene image as global feature of the landmark scene image, and extracting a SIFT descriptor as the local feature of the landmark scene image; performing initial clustering to a landmark scene image library through the global feature according to the two dimension feature of the image; selecting a plurality of representative images closest to the clustering center from each cluster to the initial clustering result through the local feature, and performing pair matching for the representative images for saliency-based geometric verification; after the geometric verification; extracting the landmark image from each cluster aggregate; and performing geometric matching and screening to the landmark image of the area interested by the user in the selected landmark scene images through the geometric verification of clusters, and gathering the landmark images with same or similar visual angles, thus realizing the mixing of the similar clusters, and extracting the landmark scene abstract.

The invention provides a method for debugging a color correction matrix. The method comprises the steps of acquiring the RGB values of color blocks on a 24-color adjustment card and storing the acquired RGB values as a target color matrix; acquiring the RGB values of color blocks on a to-be-adjusted picture and storing the acquired RGB values as a source color matrix; calculating to obtain a color correction matrix according to the target color matrix and the source color matrix; converting the color correction matrix according to the source color matrix to obtain a sample picture; according to the sample picture and the 24-color adjustment card, debugging the color correction matrix. The invention also provides an electronic device corresponding to the above method for debugging the color correction matrix. The invention also provides a method and a system for manufacturing the above 24-color adjustment card. According to the method for debugging the color correction matrix, the color correction matrix is debugged based on the 24-color adjustment card manufactured according to the manufacturing method, so that restored colors are closer to colors seen by human eyes. Therefore, the debugging effect of the color correction matrix is improved.

The invention discloses an image recognition technology-based forest fire automatic monitoring and recognition system and method, and belongs to the field of image processing. Based on the characteristics of wide application range and high economy of video monitoring image processing and mode recognition technologies, the image recognition technology-based forest fire automatic monitoring and recognition system and method are applied to the field of forest fire monitoring and early warning. The image recognition technology-based forest fire automatic monitoring and recognition method comprisesthe following steps: performing preprocessing of image noise reduction and data compression by using median filtering and gray processing methods; detecting a moving target in the monitoring image byusing a Gaussian mixture background model and a differential detection technology; extracting multi-dimensional features such as color moment, energy, entropy, inverse difference moment and circularity from the detected target image, and fusing the multi-dimensional features into a normalized feature vector; and finally, using a binary classifier of a support vector machine for identification andjudgment. According to the image recognition technology-based forest fire automatic monitoring and recognition system and method, a moving target appearing in a complex background can be well detected, and characteristic parameters are extracted for identification, and automatic fire identification is further realized, and the image recognition technology-based forest fire automatic monitoring and recognition system and method have the advantage of high monitoring accuracy.

The invention discloses a method for color standardization of pathological image based on low rank embedded non-negative matrix decomposition, the data of two pathological images are expressed in theform of observation matrix, which are converted into optical density, the low rank representation matrices of the source optical density matrix and the target optical density matrix are solved respectively, the corresponding optical density matrix is decomposed into an RGB color matrix of each dyeing component and an intensity matrix of each pixel under each dyeing component, At last, that image is convert to RGB form, so as to achieve the color standardization from the source pathological image to the target pathological image, the invention can be used for color standardization of pathological images and improvement of visualization effect, Avoiding the color inconsistency of pathological images caused by different pathological scanners, different laboratory staining methods and different staining ratio has important significance for pathological image imaging, staining separation and intelligent pathological image analysis, and has broad market prospects and application value.

The invention relates to an urban solid waste incineration process combustion condition recognition method based on flame image color feature extraction. The components of domestic urban solid waste (MSW) are complex, and domain experts usually observe flame video images according to experience to identify the combustion state of the incinerator so as to adjust the operation parameters of the MSWIprocess, so that the stable operation condition is difficult to maintain. The method comprises the following steps: firstly, carrying out defogging and denoising pretreatment on an incineration flameimage to improve image definition; then, after the image is converted into an HSV space suitable for being identified by a visual system, color moment characteristics are extracted by adopting sliding window blocks, and potential characteristics are extracted by adopting principal component analysis (PCA) so as to eliminate the colinearity among the high-dimensional color moment characteristics;and finally, taking the extracted mutually independent potential characteristics as input, and constructing a combustion condition recognition model by adopting a least square support vector machine (LSSVM) algorithm. The effectiveness of the method is verified based on actual incineration image simulation of a certain factory in China.

The invention provides a no-reference contrast distortion image quality evaluation method, which comprises the following steps: extracting color moments and information entropy features in various color spaces from a contrast distortion image, and constructing a feature set for describing image distortion; constructing a training set according to the feature set of image distortion and the prior score, and constructing a prediction model of image quality evaluation; and extracting a contrast distortion feature set of the to-be-evaluated image, and performing calculation by utilizing the imagequality evaluation prediction model to predict the image quality of the to-be-evaluated image. According to the no-reference contrast distorted image quality evaluation method provided by the invention, a multi-color space is fused, color moment and information entropy features are combined, the detection accuracy and effectiveness are well ensured, and the vacancy in the no-reference contrast distorted image quality evaluation field is filled.

The invention discloses a radio frequency fingerprint identification method based on QPSK signal bispectrum energy entropy and color moments, relating to the field of wireless communication. A bit stream signal of a sending end is subjected to QPSK mapping to obtain a signal s(n). Through up conversion, a frequency modulation signal p(n) is obtained and is inputted into a power amplifier to output a signal Phi (n), an analog signal is obtained through digital to analog conversion processing, the analog signal is sent out and is added into AWGN in a sending process, a receiving end obtains a digital signal r(n) through analog to digital conversion processing, a baseband signal is obtained through down conversion, and radio frequency fingerprint characteristics including the bispectrum energy entropy, a first-order moment and a second-order moment are extracted from the baseband signal. Then through an SVM classifier, the classification training and testing of the radio frequency fingerprint characteristics are carried out, and a test category result is obtained. Through comparing the test category result and an actual category result, a classification accuracy rate Pc is obtained. According to the method, the radio frequency signals are effectively classified, and the identification accurate rate under a low signal-to-noise ratio is improved by nearly 20% compared with a traditional method.

The invention discloses a human body behavior recognition method based on a spatio-temporal distribution map generated by motion history point clouds. The human body behavior recognition method comprises the following steps: generating an MHPC (Motion History Point Cloud); generating an STDM (Spatio-Temporal Distribution Map); extracting a color moment feature vector; extracting an LBP feature vector; training and testing a KELM classifier, and finally fusing output results by adopting a decision layers to obtain a human body action type label. The human body behavior recognition method disclosed by the invention can acquire information of human body actions under different visual angles, so that the robustness of an action angle change is improved. The STDM for expressing a human body action is more comprehensive than a depth image, and extracted features are more distinctive; extracted color moment feature and LBP feature can effectively characterize human body action types, so thatthe problem of complexity in feature extraction by using the point clouds is solved. By use of decision layer-based fusion for classification, the shortcomings of incompatibility and high dimension offeature layer fusion can be avoided.

The invention discloses an automatic detection method for the three-dimensional coordinates of an RFID (Radio Frequency Identification) tag based on DLT, which comprises the steps that an image collection platform is established; the RFID tag is collected, and an RFID tag image is obtained; pixel coordinates of the RFID tag and size pixels of the RFID tag are obtained; coordinates of the bottom ofa tag support are calculated by using a color matrix; the physical position of the vertical coordinate of the RFID tag is calculated by using a proportional algorithm; the physical position of the horizontal coordinate of the RFID tag is obtained by using a DLT algorithm; the three-dimensional coordinates of the RFID tag are obtained; and a computer drives a motor to control a conveyor belt to drive the RFID tag to move towards an RFID reader, when the RFID tag is read by an antenna of the RFID reader, a laser distance measuring sensor measures a distance value between the antenna of the RFIDreader and the RFID tag, and the distance value is stored in the computer. According to the invention, the detection result is accurate, and the detection result is not affected by the position of the RFID tag.

The present invention represents a multi-thread normalized non-negative sparse encoder based method for rapid feature representation of an image. The method comprises the specific steps of: local feature extraction of an image: densely extracting SIFT features and color moment features of each image in a data set, randomly selecting a plurality of pairs of the extracted SIFT features and color moment features to obtain a codebook by using a K-means clustering method, and optimizing solution encoder parameters according to a relaxation model based on the multi-thread normalized non-negative sparse encoder by using the codebook obtained after solution; and a test phase: using an encoder obtained in a training process to only extract the SIFT features of one input image during feature representation of the image, using the encoder to calculate coded coefficients of the SIFT features, and integrating all the coded coefficients according to a space pyramid maximized pooling manner, wherein an obtained high-dimensional feature vector is a feature vector of the image. The obtained image feature representation is further used in intelligent analysis application of classification / retrieval and the like of various images.

The invention relates to a vehicle-mounted camera-assisted distributed driving electric vehicle pavement estimation method. The method comprises the following steps: 1) respectively shooting images ofa road pavement in front of a vehicle under dry weather conditions and rainy and humid weather conditions by using a vehicle-mounted camera; 2) based on a color moment method and a gray level co-occurrence matrix method, extracting road surface image features from the acquired road surface images of the road in front of the vehicle under various weather conditions; 3) utilizing a support vector machine method to carry out classification identification on each road surface image characteristic acquired in the step 2); and 4) designing a disturbance estimator based on tire longitudinal force estimation, estimating a road adhesion coefficient, considering a mapping relationship between a road image and the adhesion coefficient, designing a fusion estimator, and further obtaining a fusion estimation result. Compared with the prior art, the method has the advantages of high estimation precision, good real-time performance, strong robustness and the like.

The embodiment of the invention provides a method and device for processing a picture in a night mode. The method for processing the picture in the night mode comprises the steps of obtaining a to-be-processed picture when a browser is in the night mode and decomposing the to-be-processed picture into a plurality of to-be-processed sub-pictures with preset sizes; processing the to-be-processed sub-pictures according to a preset color matrix corresponding to the night node to obtain target sub-pictures; and merging the target sub-pictures and obtaining a target picture of the to-be-processed picture in the night mode. According to the method and the device provided by the invention, the picture of a webpage in the night mode is processed after the browser is switched into the night node, so that the contrast of the brightness of the target picture and the brightness of surrounding environment light is relatively small.

The invention discloses a black smoke vehicle detection method based on a self-organizing background difference model and a multi-feature fusion, which comprises the following steps: detecting a moving object from a video surveillance by using the self-organizing background difference model, and determining a key area; transforming the key region image into YCrCb color space and extracting the color moment feature; transforming the key regions into gray space, and extracting the local ternary mode histogram and edge direction histogram; according to the position of the key region of the current frame, extracting the corresponding regions of several frames from the whole frame sequence, concatenating the same features extracted from all the temporal regions to form the feature vectors of each class, and normalizing the feature vectors of each class to form the final feature vectors; classifying the final eigenvectors by using the pruned radial basis function neural network classifier; identifying the key areas of smoke and further identifying the smoky vehicles. The invention can further improve the identification rate, reduce the false alarm rate, and has good identification effecton smoky vehicles with relatively light black smoke.

The invention discloses a silkworm cocoon classification method based on color features and a support vector machine. The silkworm cocoon classification method comprises the following steps: collecting silkworm cocoon images; Preprocessing the silkworm cocoon image; extracting color moments and color histogram features of the silkworm cocoon images; Analyzing color feature main components of the silkworm cocoon image; and establishing a silkworm cocoon classifier model by using a support vector machine, and identifying the types of the new silkworm cocoon images by using a classifier. According to the invention, the silkworm cocoon image is subjected to color histogram and color moment information extraction; and classification is carried out by combining principal component analysis dimensionality reduction with a support vector machine, so that the problem that a production inspection mechanism cannot automatically identify and classify four secondary cocoons which are yellow cocoons, rotten cocoons, heading cocoons and thin-skinned cocoons and seriously influence production and inspection at present is effectively solved, and positive significance is achieved for improving the production quality of textiles.

The invention discloses an electronic component positioning and detecting method based on SURF feature matching. The method comprises the following steps: acquiring an image of a sample circuit boardand an image of a circuit board to be tested, establishing a coordinate system for an original sample image and an original image to be tested, and performing mean downsampling on the two images respectively to obtain a reference sample image and a reference image to be tested; sURF feature point extraction and feature point matching are performed on the reference sample image and the reference to-be-detected image respectively; obtaining a matching point, calculating a geometric transformation relationship between the reference sample image and the reference to-be-detected image, positioningcoordinates of the electronic component in the original to-be-detected image, and judging whether the electronic component in the original to-be-detected image has defects or not. According to the invention, the electronic component on the circuit board to be detected is positioned after down-sampling is carried out on the image, and the defect of the electronic component is detected according tothe color moment difference, so that the positioning time is greatly shortened, the image data processing complexity is reduced, and the detection accuracy and the detection precision are improved.

The invention relates to a highway tunnel fire smoke dynamic detection method, and belongs to the technical field of highway tunnels. The method comprises the following steps: S1, extracting a motion area; S2, segmenting a suspected smoke region; S3, feature analysis and extraction; and S4, fire smoke detection. According to the method, suspected smog area segmentation is carried out in a YUV color space through a color filtering rule based on a motion area extracted by a Gaussian mixture model, then color and texture features of smog in the suspected smog area are mainly researched, a color moment of the smog is calculated, a uniform local binary pattern (ULBP) and a gray level co-occurrence matrix (GLCM) of the smog are extracted to carry out feature fusion, a machine learning classifier is used for model training to distinguish smoke and non-smoke, smoke recognition is completed, and an optimal algorithm is selected according to an analysis result.

The invention discloses a color correction method and a related device thereof, wherein the method comprises the steps of obtaining color information and a color temperature estimation value of to-be-debugged equipment; selecting corresponding color data from a color space data table of reference equipment according to the color information and the color temperature estimation value; obtaining a target color matrix by using the color data, and obtaining a source color matrix; and calculating a color correction value according to the target color matrix and the source color matrix, and correcting the color information of the to-be-debugged equipment by using the color correction value. Through the method, the color style debugging efficiency of the equipment is effectively improved.

The invention discloses an image deduplication method and device based on feature point matching, and the method comprises the steps: extracting the first-order color moment and second-order color moment features of each image in a redundant image set to obtain a color feature vector; clustering the color features by adopting a clustering algorithm, and clustering the images into K classes; extracting SURF features in all the images; carrying out SURF feature point matching on the images in each class, if the number of matching points is larger than a set matching threshold value, judging thatthe images are the same, and deleting redundant images; repeating for multiple times to obtain a deduplicated image data set. According to the method, through pre-classification, the defect of greatcalculation complexity caused by pairwise comparison required by feature point matching is improved.

The invention discloses an image color correction method and device, and belongs to the technical field of image processing, and the method comprises the steps: reading a shot color card picture, recognizing the color of the shot color card picture, and generating a color matrix A; acquiring a color matrix B generated by the color of the standard colorimetric card; calculating a color correction matrix X according to the color matrix A and the color matrix B; and inputting the color correction matrix X into a camera, and correcting the image by the camera in an image processing process. The method is good in automation effect and high in image correction efficiency.

The invention is suitable for the field of video investigation, and provides a quick image search method for video investigation. The quick image search method comprises the following steps: obtaining a background model; obtaining a moving target; obtaining a candidate target matching pair, and completing the tracking processing of the moving target; determining the information amount of the picture according to the probability of each pixel in a target image, and taking the target image corresponding to the maximum value of the information amount of the picture as the output of an optimal target image; calculating the characteristics of the optimal target image, and carrying out dimensionality reduction on characteristic vectors according to a compressive sensing theory, wherein the characteristics include HSV column diagram, color autocorrelogram, color moment, small wave torque and gradient direction column diagram,; calculating the similarity of sample pictures and the optimal target image according to the characteristic vectors of the image, and carrying out target ranking on the sample pictures according to the similarity so as to obtain the search result. By providing a novel moving target optimal picture selection strategy, the information loss caused by tracking problem and invalid picture selection is further reduced.

The invention discloses a black and odorous water body image extraction and recognition method. The method comprises the following steps: A, data preprocessing; dividing the image into three color channels, converting a pixel value matrix, and extracting an image of 101 * 101 pixels in the center of the water sample image; B, image feature extraction: extracting color moments of each order; C, taking 80% of samples as training samples, and taking the rest 20% of samples as test samples; D, training the model by using a training set sample; E, evaluating the performance of the model by using the test sample; the black and odorous water body image is obtained through the technologies of sampling, photographing and optical fiber transmission of the black and odorous water body, the image is identified through machine learning and other methods, the black and odorous water body is classified and registered, the black and odorous water body can be rapidly identified and the black and odorous water body treatment condition can be monitored in a region range, time and labor are saved, and application and popularization are easy.

The invention provides a color difference guided convolution point cloud semantic segmentation method. The method comprises the steps of converting colored point cloud RGB into HSV; for the point cloud under the current density, solving a k neighborhood to obtain a relative position y of a central point and a neighborhood point, and sending the relative position y to an MLP to obtain a feature FG;obtaining d1, d2 and d3 in three directions according to neighborhood different channel color moment sorting and y, and selecting corresponding features of nearby points; convolution is performed onthe selected features in three directions and color channels, so that features FC can be obtained; cascading the FG with the FC to obtain a global feature F; performing down-sampling, and repeating the steps 2-5; f and FC interpolations under adjacent densities are recovered, and a prediction result of each point is obtained; based on the cross entropy loss function, minimizing the loss function through gradient descent, and training neural network parameters; and after the parameters are trained, when a new to-be-segmented point is given, executing the steps 1-8 to obtain a segmentation result. It can be seen from experiment results that the method can obviously improve the point cloud semantic segmentation precision under various types, and is suitable for indoor and outdoor scenes.

The invention relates to a solid waste incineration condition recognition method based on multi-scale color moment characteristics and a random forest. Redundancy and complexity of incineration flameimage features increase the difficulty of urban solid waste incineration (MSWI) combustion condition recognition. Complexity of solid waste components and inherent nonlinearity, time-varying property,uncertainty and the like of a solid waste incineration process cause instability of incineration image feature distribution. A traditional method based on a fixed sliding window can only extract fixed size features and cannot reflect global and local features, and the working condition recognition accuracy is reduced. The method comprises the following steps: firstly, performing defogging and denoising preprocessing on an image; then, using a sliding window based on a priori set scale to extract color moment features of different scales of the flame image; finally, using the classification precision as a criterion function, using a random forest (RF) algorithm based on feature selection, to achieve accurate recognition of the MSWI incineration working condition. An experiment result verifies the effectiveness of the method.

The invention provides an aerial photography water body extraction method and system based on deep learning. The method comprises the steps of: collecting A multispectral image, obtaining an RGB imagethrough true color synthesis, and obtaining a hue image and a gray level image through color space conversion; obtaining a water body characteristic graph F1 according to the water body characteristic index; obtaining an energy texture feature map and an entropy texture feature map of the grayscale image according to the grayscale co-occurrence matrix, and obtaining a fine-grained feature map F2of the water body in combination with the tone map; inputting the RGB image, the F1 and the F2 into a trained semantic extraction encoder to obtain a feature map F3; and calculating a plurality of color moments of the RGB image, inputting the color moments into the FC for full connection, outputting a plurality of neurons, multiplying the neurons by the corresponding feature maps, finally extracting water body features through a decoder, and outputting a water body segmentation map.

The invention relates to a bar code image restoration method and device, computer equipment and a storage medium, and the method comprises the steps of carrying out the correction processing of a bar code image collected by terminal equipment, and obtaining a corrected bar code image, and taking the corrected bar code image as a to-be-recognized bar code image; identifying the color of each pixel in the to-be-identified barcode image line by line, and generating a color matrix corresponding to the to-be-identified barcode image according to a color identification result; determining a color identifier corresponding to each column of the color matrix according to the number of black elements in each column of the color matrix; generating a target barcode image corresponding to the color matrix according to the color identifier. The invention comprises performing color recognition on a corrected to-be-recognized barcode image to generate a color matrix matched with the image; determining the color identification of each column according to the number of the black elements of each column in the color matrix, and forming the corresponding target barcode image according to the color identification, so that the barcode image is repaired, and the barcode image recognition efficiency is improved.

The invention relates to a target classification method for logistic regression based on principal component analysis, and belongs to the field of picture processing and pattern recognition. The method is characterized by comprising the following steps: (1) determining color moment characteristics; (2) determining texture feature parameters; (3) determining a characteristic parameter correlation coefficient matrix; (4) determining an input feature vector; and (5) taking the feature vectors obtained in the previous step as input of a logistic regression model, and establishing the logistic regression model for training to realize classification of different training samples. The feature vectors of the sample set are subjected to dimensionality reduction by using a principal component analysis method; the classification method is used as an input feature vector of logistic regression to carry out model training, a complex algorithm process is not needed to realize target classification,the classification method can improve the stability and high efficiency of the model on the basis of ensuring the classification accuracy, and an accurate and stable classification method is provided.

The invention relates to a no-reference tone mapping image quality evaluation method, and the method comprises the following steps: (1), carrying out the global feature extraction: calculating a colormoment feature, and recording the color moment feature as f1; converting the to-be-evaluated image I into a grayscale image, respectively counting the proportions of all pixels falling into a dark area and a bright area of one image, and finally representing the light and shade distribution characteristics f2 of the image by using the dark area ratio and the bright area ratio; Wherein the globalentropy feature is recorded as f3; (2) performing local feature extraction: dividing the image I into 16 * 16 local blocks in a space domain, calculating the sum-difference ratio of the maximum pixelvalue Imax to the minimum pixel value Imin to the maximum pixel value B in the blocks, and recording the sum-difference ratio as a contrast ratio f4; calculating the local entropy of the image block and recording the local entropy as f5; dividing the image I into 16 * 16 blocks in a wavelet domain, uniformly expressing wavelet decomposition coefficients in four directions of approximation, horizontal, vertical and diagonal by C, carrying out energy operation on the coefficients C, and recording the wavelet energy as f6; (3) quality assessment.

The invention discloses an image recognition method and device and a storage medium. The method relates to the technical field of image processing, and comprises the steps of carrying out the grayingprocessing, feature vectorization processing and sample set training of a normalized standard seal image, obtaining a trained sample set, carrying out the classification of the trained sample set according to a preset rule, and obtaining a classification recognition model; and enabling the classification and identification model to identify the normalized to-be-detected seal image, and calculatingthe matching degree of the standard seal image and the to-be-detected seal image by adopting a Match algorithm according to the color matrix. The technical scheme of the invention is provided for identifying the seal. The seals are with similarity, so the model can be conveniently trained, the trained model is high in recognition degree; experience data does not need to be obtained through a large number of experiments or through an expert system technology, time is saved, the technical scheme can replace manpower to identify the authenticity of the seal, the process speed of target image recognition is relatively high, the technology is simple, and the recognition accuracy is high.