174 results about "Salient objects" patented technology

The invention discloses a deep learning-based weakly supervised salient object detection method and system. The method comprises the steps of generating salient images of all training images by utilizing an unsupervised saliency detection method; by taking the salient images and corresponding image-level type labels as noisy supervision information of initial iteration, training a multi-task fullconvolutional neural network, and after the training process is converged, generating a new type activation image and a salient object prediction image; adjusting the type activation image and the salient object prediction image by utilizing a conditional random field model; updating saliency labeling information for next iteration by utilizing a label updating policy; performing the training process by multi-time iteration until a stop condition is met; and performing general training on a data set comprising unknown types of images to obtain a final model. According to the detection method and system, noise information is automatically eliminated in an optimization process, and a good prediction effect can be achieved by only using image-level labeling information, so that a complex andlong-time pixel-level manual labeling process is avoided.

Described is a bio-inspired vision system for attention and object segmentation capable of computing attention for a natural scene, attending to regions in a scene in their rank of saliency, and extracting the boundary of an attended proto-object based on feature contours to segment the attended object. The attention module can work in both a bottom-up and a top-down mode, the latter allowing for directed searches for specific targets. The region growing module allows for object segmentation that has been shown to work under a variety of natural scenes that would be problematic for traditional object segmentation algorithms. The system can perform at multiple scales of object extraction and possesses a high degree of automation. Lastly, the system can be used by itself for stand-alone searching for salient objects in a scene, or as the front-end of an object recognition and online labeling system.

The invention discloses a deep learning saliency detection method based on the global a priori and local context. The method includes the steps of firstly, performing superpixel segmentation for a color image and a depth image, obtaining a global a priori feature map of each superpixel based on middle-level features such as compactness, uniqueness and background of each superpixel, and further obtaining a global a priori saliency map through a deep learning model; then, combining the global a priori saliency map and the local context information in the color image and the depth image, and obtaining an initial saliency map through the deep learning model; and finally, optimizing the initial saliency map based on spatial consistency and appearance similarity to obtain a final saliency map. The method of the invention can be used for solving the problem that a traditional saliency detection method cannot effectively detect a salient object in a complex background image and also for solving the problem that a conventional saliency detection method based on deep learning leads to false detection due to the presence of noise in the extracted high-level features.

The invention discloses a self-adaptive display method for an image on display terminals with different sizes, comprising the following steps of: (1) constructing a visual attention model to obtain a final saliency map; (2) marking a most salient position: dividing the obtained final saliency map into blocks, marking the block with the maximum salient energy as the most salient position in the saliency map; and (3) self-adaption regulating: gradually expanding a display window towards x and y positive and negative coordinate axes according to the size and the proportion of a target display by taking the center of the marked maximum salient position as the original point of the coordinate axes. With the method, the easily occurred problem of loss of important attention regions of the image when the image is displayed on displays with different sizes is avoided, in addition, the distortion of the important regions of the image, i.e. the regions which an audience pay most attention to, is lowered to the minimum, thereby keeping a salient object in the display content accurate and complete.

The invention discloses a high-resolution remote sensing image feature matching method. The method comprises the steps of extracting salient object regions on a reference image; extracting the SIFT feature points of the reference image and the SIFT feature points of an input image; searching for a plurality of candidate matching points for each SIFT feature point of the reference image from a SIFT feature point set of the input image; searching for optimal matching points for a SIFT feature point set in each salient object region of the reference image from a candidate matching point set composed of corresponding candidate matching points; obtaining an optimal matching point set of each salient object region, and using the union set of SIFT feature optimal matching point sets corresponding to each salient object region as a final feature matching set. According to the high-resolution remote sensing image feature matching method, matching is carried out based on the SIFT features in the salient object regions, a large amount of redundant information is filtered, and matching performance is improved; the number of exterior points is reduced, and robustness to change of view and image deformation is improved. The high-resolution remote sensing image feature matching method can be widely used in multiple application systems for image registration and three-dimensional reconstruction.

The invention belongs to the technical field of remote sensing image target detection, and in particular relates to a multi-spectral image ship target rapidly detection method based on a selective visual attention mechanism. The method comprises the following steps: constructing a multi-dimensional database into a biquaternion form; and extracting a salient region for detecting a salient object by use of the phase spectrum of biquaternion Fourier transform, thereby achieving integrated treatment of the multi-dimensional data and fully utilizing the information of the spectrum domain and the space domain. By utilizing the method, the defects of the conventional multi-spectral image ship detection method, such as high calculation complexity and complex parameter setting can be effectively overcome. The test results of the analog data and real multi-spectral remote sensing data show that the method has the advantages of good detection result and stronger robustness to noises. The method has great application value in fields of the marine fishery, marine transportation regulation, marine military monitoring and the like.

The invention provides an infrared and visible light image fusion method based on salient objects. The infrared and visible light image fusion method based on the salient objects includes following steps: building nonlinear scale space representation for an infrared image and a visible light image which respectively comprise a plurality of objects in a given scene; using a visual attention computational model to compute visual attention salient maps of the infrared image and the visible light image based on the nonlinear scale space representation of the images; using a return inhibition mechanism to select salient object areas from the infrared image and the visible light image based on the visual attention salient maps of the infrared image and the visible light image, and computing all salient object areas in the whole scene; performing rectification operation on the infrared image and the visible light image, using a pixel level fusion algorithm to perform fusion treatment on the salient object areas, and using a feature level fusion algorithm to perform fusion treatment on non-salient object areas; generating a fusion image of the infrared image and the visible light image of the whole scene by synthesizing results.

The invention discloses a salient object-based image retrieval method and system. The method comprises the steps of performing saliency detection on a query image containing a salient object to determine a region where the salient object of the query image is located; determining visual features of the region where the salient object of the query image is located; determining a semantic type of the salient object of the query image; performing similarity measurement on the visual features of the salient object of the query image and visual features of salient objects of images with the same semantic type in an image library, and determining the images, meeting a condition that the similarity between the images and the query image is greater than a similarity threshold, in the image library. According to the method and the system, the image retrieval is carried out through the visual features of the region where the salient object of the image is located, so that the background interference is avoided; by determining the semantic type of the salient object of the query image, the images with different semantic types in the image library are filtered, so that the semantic gap of the image retrieval is reduced, the image retrieval complexity is lowered, and the image retrieval accuracy is further improved.

The invention discloses a salient target detection method based on a cascade convolutional network and adversarial learning. The method comprises the following steps: 1, designing a global saliency estimator E; 2, designing a local saliency refiner R; 3, combining the global saliency estimator E and the local saliency refiner R into a generator G based on a cascade convolutional neural network forgenerating a saliency map; 4, optimizing the generator G; 5, designing an adversarial learning discriminator D to distinguish a real saliency map from a predicted saliency map generated by a generator G; and 6, the generator G and the adversarial learning discriminator D follow the CGAN strategy and are trained in a complete end-to-end manner, so that the generator G can better understand the structure information of the salient object, and a good saliency detection result is obtained. According to the method, the structural information is learned implicitly through confrontation learning, sothat significance target detection can be well carried out, and a best result is obtained on a plurality of databases.

The invention relates to a painting style migration method based on saliency matching. The method is characterized by providing a painting style migration depth neural network model based on saliencymatching; and the model is formed by four modules: feature extraction, sub painting style migration, saliency-based region decomposition, and painting style image synthesis. During training, branch training is carried out on the constructed painting style migration network model and thus each branch is optimized towards target effects. Compared with the prior art, the painting style migration method has the following advantages: firstly, painting style migration is carried out based on the consistency of object saliency in a content image and sub painting style saliency in a painting style image and thus the salient painting style can be migrated to the salient object in the content image, thereby ensuring unchanging of the object saliency sequence in the content image after style migration; and objects with different saliency carry on the single sub painting style, so that chaotic style mixing caused by carrying on different sub painting styles by the same object is avoided; and secondly, the generated image with the painting style becomes smooth without noises.

The invention discloses a salient object detection method based on video. The method comprises the following steps: performing super-pixel segmentation on a picture, and utilizing optical flow information and color information to obtain a preliminary salient object detection area; acquiring picture background information by using gradient information; fusing the preliminary object detection area and pruning the background to obtain a final salient object zone. According to the method, the similarity of pixels is detected by combining statistical motion information (i.e. the optical flow information) and static information (i.e. the color and gradient information) on the pixel level, foreground and background cues are fused in the result, and the purpose of detecting a salient object in video is finally fulfilled.

An adaptive video presentation method for automatically presenting a video with stream-embed information based on content analysis of the video on a smaller display with a limited screen size is provided. The method comprises steps of determining a salient object group containing at least one salient object based on perceptual interest value of macroblocks for each frame of said video, extracting a window having a minimum size containing the salient object group for a scene of the video, characterized in that it further comprises steps of comparing size of the extracted window with the smaller display size; and presenting at least a selected area of the extracted window containing at least a part of the salient object group for the scene on the smaller display in different operation modes based on the result of the comparison steps for different motion mode for the scene of the video.

The invention discloses a salient object segmenting method in a stereo image based on depth information and image cutting. The method comprises: inputting and segmenting an original image and a depth map in order to acquire a plurality of areas; computing a saliency map of the original image in combination with area-class depths, colors, and spatial domain information; acquiring a threshold value of the saliency map obtained through computation in order to complete initial segmentation of the original image, thereby acquiring an object/background seed point; constructing a map by using the depth map, the saliency map obtained through computation, and a salient weighted histogram, and designing a cost function; and completing salient object segmentation at one time by using a maximum flow minimum cut algorithm. The method reasonably utilizes the depth information and the saliency map, and more accurately and automatically segment the salient object in the stereo image.

Methods, systems, and non-transitory computer readable storage media are disclosed for generating focused preview images that include the subjects of interest of digital images. For example, in one or more embodiments, the disclosed system utilizes a machine-learning model to generate a saliency map for a digital image to indicate one or more salient objects portrayed within the digital image. Additionally, in one or more embodiments, the system identifies a focus region based on focus information captured by a camera device at the time of capturing the digital image. Furthermore, the system can then utilize the saliency map and the focus region to generate a focused preview image. For instance, in one or more embodiments, the system crops the digital image based on an overlapping portion of the saliency map and the focus region to generate a focused preview image.

There is provided a moving image extracting apparatus including a movement detecting unit which detects movement of an imaging apparatus at the time when imaging a moving image based on the moving image imaged by the imaging apparatus, an object detecting unit which detects an object from the moving image, a salient object selecting unit which selects an object detected by the object detecting unit over a period of predetermined length or longer as a salient object within a segment in which movement of the imaging apparatus is detected by the movement detecting unit, and an extracting unit which extracts a segment including the salient object selected by the salient object selecting unit from the moving image.

The invention discloses a saliency detection method based on a conditional random field. Saliency detection is considered to be an image annotation problem. Multi-scale comparison is used, and salient feature graphs are obtained through three different saliency calculation modes including a center-periphery histogram and color and space distribution. The weight of saliency of each salient feature graph is calculated through CRF study, and a model parameter is obtained through a maximum likelihood estimation method to estimate the optimal solution. Finally, the CRF is used for detecting and testing images. The method can detect salient objects more precisely, results obtained through detection are high in resolution ratio, an object boarder is defined precisely, and the method is little in calculation complexity.

The invention discloses an image description generation method based on an architectural short sentence constraint vector and a dual visual attention mechanism, A semantic model for automatically describing the visual content information of images is obtained by training a large number of labeled texts. The semantic model is composed of three parts: a short sentence generation model, a dual visualattention mechanism and a constrained language model. The text description can be generated automatically for any input test image. The invention can effectively establish the relationship between the word and the image in the text description, and has good performance for describing the salient objects or scenes of the image based on semantics.

The invention provides a matrix factorization-based hyperspectral image saliency target detection method. According to the method, the spectral gradients of an original hyperspectral image are calculated in spectral dimension, the spectral gradient features of the image are extracted, and therefore, adverse effects caused by illumination are eliminated, and at the same time, an image feature matrix is constructed; matrix low-rank sparse decomposition is performed on the image feature matrix, so that a low-rank matrix corresponding to a background part and a sparse matrix corresponding to a saliency target are obtained; and therefore, the problem of nonuniformity of block division in a salient object can be solved, and saliency target detection is realized with decreased computational complexity.

The invention provides a mobile visual focus based image vision salient detection method. The mobile visual focus based image vision salient detection method includes inputting an original image, applying an image based partitioning algorithm to partitioning the original image into K areas, performing quantization and high-frequency color screen on the original image, using a partition image to obtain partition areas corresponding to a quantization image, performing area comparison and calculation to obtain a salient value of some area so as to obtain an initial salient image, using an image center as an initial visual focus to perform weighting calculation on the initial salient image, using a salient attraction module to calculate a moving distance and a moving direction of a visual focus so as to obtain the next visual focus until the current focus distance and the last focus distance are smaller than 1, and using a final visual focus to weight a salient image obtained by a visual focus prior to the final visual focus so as to obtain a final salient image. Compared with image vision salient detection methods in the prior art, the mobile visual focus based image vision salient detection method has the advantages of being high in accuracy of the salient image, and capable of evenly highlighting salient objects and well inhibiting image backgrounds.

The invention discloses an RGB-D image salient object detection method based on salient center prior. The RGB-D image salient object detection method comprises steps that salient center prior based on a depth map and salient center prior based on an RGB image are provided; according to the salient center prior based on the depth map, Euclidean distances between other super pixels in the RGB image and the depth characteristics of the center super pixels of the depth map salient object are calculated, and are used as the salient detection results of the salient weight strengthened RGB image, and therefore the RGB image salient detection is effectively guided by the depth characteristics, and the RGB image salient detection results are improved; according to the salient center prior based on the RGB image, the Euclidean distances between the other super pixels of the depth map and the CIELab color characteristics of the center super pixels of the RGB image salient object are calculated, and are used as the salient detection results of the salient weight strengthened depth map, and therefore the depth map salient detection is effectively guided by the RGB characteristics, and the depth map salient detection results are improved.

Embodiments of the present disclosure include a computer-implemented method that receives a digital image input, the digital image input containing one or more dynamic salient objects arranged over a background. The method also includes performing a tracking operation, the tracking operation identifying the dynamic salient object over one or more frames of the digital image input as the dynamic salient object moves over the background. The method further includes performing a clustering operation, in parallel with the tracking operation, on the digital image input, the clustering operation identifying boundary conditions of the dynamic salient object. Additionally, the method includes combining a first output from the tracking operation and a second output from the clustering operation to generate a third output. The method further includes performing a segmentation operation on the third output, the segmentation operation extracting the dynamic salient object from the digital image input.