Thomson Licensing

Siemens Healthcare Diagnostics Gmbh

Lenovo (beijing) Limited

Interdigital Vc Holdings, Inc.

Interdigital Madison Patent Holdings, Sas

Jiang Tian

Kangying Cai

Wenfei Jiang

Tao Luo

Terrence Chen

Zhi Zhang

Yao Zhang

Shanhui Sun

Benjamin S Pollack

Yao-Jen Chang

Stefan Kluckner

Juan Xu

Wei Liang Meng

Weiliang Meng

Zhi Yuan Zhang

An image processing method includes obtaining a first quantity of to-be-analyzed images and performing fusion and enhancement processing on the first quantity of to-be-analyzed images through an image analysis model to obtain a first target image. Each to-be-analyzed image corresponds to a different target modality of a target imaging object. The first target image is used to enhance display of a distribution area of an analysis object of the first quantity of to-be-analyzed images. The analysis object belongs to the imaging object. The image analysis model is obtained by training a second quantity of sample images corresponding to different sample modalities. The first quantity is less than or equal to the second quantity. The target modality belongs to the sample modalities.

Multi-modality image processing method and device

A method of characterizing a serum and plasma portion of a specimen in regions occluded by one or more labels. The characterization may be used for Hemolysis, Icterus, and/or Lipemia, or Normal detection. The method captures one or more images of a labeled specimen container including a serum or plasma portion, processes the one or more images to provide segmentation data and identification of a label-containing region, and classifying the label-containing region with a convolutional neural network (CNN) to provide a pixel-by-pixel (or patch-by-patch) characterization of the label thickness count, which may be used to adjust intensities of regions of a serum or plasma portion having label occlusion. Optionally, the CNN can characterize the label-containing region as one of multiple pre-defined label configurations. Quality check modules and specimen testing apparatus adapted to carry out the method are described, as are other aspects.

Methods and apparatus for determining label count during specimen characterization

In the present invention are a method and apparatus for processing a block to be processed of a urine sediment image. The method comprises: dividing a block to be processed into a plurality of grids; calculating an n-dimensional local feature vector of each grid of the plurality of grids, where n is a positive integer; in the block to be processed, merging at least two adjacent grids of the plurality of grids into an intermediate block; calculating an intermediate block merging feature vector of the intermediate block; according to a predetermined combination rule, combining the intermediate block merging feature vectors obtained for different intermediate blocks of the block to be processed into a general combination feature vector of the block to be processed; and by way of taking the general combination feature vector as a feature in a feature set of block processing, processing the block to be processed.

Method and apparatus for processing block to be processed of urine sediment image

A 3D model can be modeled using 'pattern-instance?representation. To describe the vertices and triangles, properties of the instance, for example, texture, color, and normal, are adjusted to correspond to the order in the pattern. The texture of an instance is encoded depending on its similarity with the texture of a corresponding pattern. When instance texture is identical or almost identical to the pattern texture, the instance texture is not encoded and the pattern texture will be used to reconstruct the instance texture. When the instance texture is similar to the pattern texture, the instance texture is predictively encoded from the pattern texture, that is, the difference between the instance texture and pattern texture is encoded, and the instance texture is determined as a combination of the pattern texture and the difference.

Method and apparatus for compressing texture information of three-dimensional (3D) models

A 3D model can be modeled using 'pattern-instance' representation, wherein an instance component may be represented as transformation (for example, rotation, translation, and scaling) of a pattern. To improve compression efficiency, the quantization parameters for the rotation part and translation part for transformation of an instance can be determined based on the quantization parameter used for encoding a corresponding pattern. Specifically, the quantization parameter for the rotation part may depend on the size of the instance, and the quantization parameter for the translation part may depend on the scale of translation. That is, a larger instance may use a finer quantization parameter for the rotation part. The quantization parameters are so determined that quantization errors caused by compressing the patterns, the translation part of transformation, and the rotation part of transformation are at similar levels.

Method and apparatus for repetitive structure discovery based 3D model compression

A method and apparatus for generating a bitstream representative of a 3D model, and a method and an apparatus for processing the same. A 3D model is modeled by using a using a 'pattern-instance' representation, wherein a pattern is a representative geometry of a repetitive structure, and the connected components belonging to the repetitive structure is call an instance of the corresponding pattern. After discovery of the repetitive structures and their transformations and properties, the present embodiments provide for generating a bitstream in either a first format or a second format. In the first format, the pattern ID and its associated transformation and property information are grouped together in the bitstream, and in the second format the pattern ID, transformation property and property information are grouped together according to information type.

Method and apparatus for generating a bitstream of repetitive structure discovery based 3D model compression

A 3D model can be modeled using 'pattern-instance' representation, wherein an instance component may be represented as transformation (for example, rotation, translation, and scaling) of a pattern. Quantization errors may be introduced when encoding rotation information, causing different vertex coordinate errors at different 5 vertices of an instance. To efficiently compensate the vertex coordinate errors, the encoder decides a quantization parameter for compensating a vertex coordinate error. The quantization parameter is signaled in the bitstream as a quantization index. The quantization index, a quantization table the indicates a mapping between quantization indices and quantization parameters, and vertex coordinate errors are 10 encoded into a bitstream. The quantization table may be built based on statistical data. At the decoder, the vertex coordinate error is decoded based on a quantization parameter, which is determined from a received quantization index.

Method and apparatus for vertex error correction

It is provided a method for constructing a graph representation for a 3D object, wherein comprising the steps of generating an initial graph representation for the 3D object, wherein a node in the initial graph representation corresponds to a component of the 3D object and a symmetric indication value uniquely identifying a symmetric group is associated with a symmetric node; selecting two nodes from one symmetric group that has the most symmetric nodes, wherein the distance between the two nodes are the largest among distance between any other two nodes within the symmetric group; obtaining two set of nodes by expanding separately from the two nodes to their directly connected nodes, wherein if a node is connected directly to both two nodes, the node is excluded from the two set of nodes; and if determining that the two sets of nodes match, updating the graph representation by grouping each set of nodes into one node.

Method and device for constructing graph representation for a 3D object

Disclosed are a method and apparatus for processing a 3D model. To preserve fine structures while de-noising a 3D mesh model, the local structural information around a vertex is captured when designing a de-noising filter. In particular, for a current vertex to be processed, a path, for example, a geodesic path, is determined between the current vertex and each neighboring vertex. For each mesh edge along the path, local variations are calculated for the two end vertices of the mesh edge using a covariance matrix, and a geometric variation for the mesh edge is calculated as the difference between the two local variations. Then structural information for the region between the current vertex and a neighboring vertex is calculated as a function of the geometric variations for mesh edges along the path, for example, as the maximum geometric variation along the path. The present principles can also be adjusted to be used in de-noising 3D point-based models.

Method and apparatus for 3D mesh de-noising

To reduce the entropy of occupancy codes of an octree and improve compression efficiency, the present principles provide a method and an apparatus for traversing sub-cells in a cell according to the geometrical property of 3D models. That is, a surface smoothness measure is calculated for each sub-cell and the sub-cells are traversed in a pre-determined order of surface smoothness measures. To compute the surface smoothness measure, a sub-cell is connected to neighboring cells of its parent cell to form triangles and angles. Subsequently, the triangle areas or angular measures are used to compute the surface smoothness measure. When the connectivity information is not available in the 3D model data, the present principles also provide a method and an apparatus for estimating the connectivity.

Efficient compression of 3D models based on octree decomposition

A 3D model can be modeled using pattern-instance representation, wherein an instance component may be represented as transformation (for example, rotation, translation, and scaling) of a pattern. Quantization errors may be introduced when encoding rotation information, causing different vertex coordinate errors at different vertices of an instance. To efficiently compensate the vertex coordinate errors, an upper bound can be estimated for the vertex coordinate error of a vertex. Based on the upper bound, the codec decides whether the vertex coordinate error of the vertex needs to be compensated, and decides a quantization parameter for compensating the vertex coordinate error if compensation is needed. The upper bound can be estimated at both the encoder and decoder, and thus, no explicit signaling is needed to indicate whether vertex coordinate error compensation is used or to indicate the quantization parameter for the vertex coordinate error.

Vextex correction method and apparatus for rotated three-dimensional (3D) components

Many 3D mesh models have a large number of small connected components that are repeated in various positions, scales and orientations. The respective positions are defined by the position of at least one reference point per component. For an enhanced encoding of the positions of the respective reference points, a given space is divided into segments and the number of points lying in each particular segment is determined. When a cell with at least n points is subdivided into child cells, an indication is added indicating if all points of a parent are in only one child cell. If so, the index of the only non-empty child node is encoded, while otherwise the number of points in one of the two child cells is decremented and encoded. The invention avoids non-effective subdivisions of a cell, and therefore improves the compression efficiency.

Growing community of inventors

Beijing, China

Jiang Tian