31results about How to "Increase blur" patented technology

The invention discloses a semantic fuzzy search method based on a sentence-level deep learning language model. In the present invention, the present invention has a high degree of ambiguity, and the present invention introduces a deep learning language model, which fully considers semantic issues, and can retrieve sentences with high semantic similarity with the target sentence, and use the method of hierarchically calculating semantic similarity Efficiently judge the semantic similarity between sentences; the operation speed is fast, and the vectorized processing is used instead of the conventional loop traversal to process text, ensuring that each semantic matching task unit can be processed in parallel, which greatly improves the search speed; The search recall rate is high, and the use of implication index makes the system more robust to grammatical interference, effectively improving the search recall rate; the system is flexible, and the present invention integrates mechanisms such as semantic understanding, fuzzy query, and precise information positioning , and then encapsulate the entire algorithm module with an interface, which is convenient for users to call.

A range discriminating optical sensor for use in a vehicle safety system is provided. The range discriminating sensor is designed to create a small depth of field such that objects at a range outside of the depth of field are eliminated from the image. Generally, the optical sensor uses a large aperture to limit the depth of field. An electronic imaging element is used to detect the image projected by the lens. The electronic imaging element may have a small pixel size which further reduces the depth of field. A signal processor can be used to determine the distance from the image sensor with objects with the selected depth of field appearing in focus and those outside of the depth of field appearing out of focus.

In an ultrasonic image processing apparatus, an original image in the form of a three-dimensional ultrasonic image is generated from volume data based on the volume rendering method. Directional interpolation processing is then applied to the original image. More specifically, a corresponding point corresponding to a noted pixel in the display image is first determined on the original image. Then, on the original image, a plurality of candidate directions extending through the corresponding point are set, and a dispersion value is computed for each of the candidate directions. Further, a normal direction with the maximum dispersion value is specified among the candidate directions, and a reference direction along the contour of a tissue is determined in the direction orthogonal to the specified normal line. Alternatively, the reference direction is determined as a direction with the minimum dispersion value among the plurality of candidate directions. Based on a plurality of pixel values existing in the reference directions, an interpolated pixel value of the noted pixel is computed. Contrast emphasis processing is further applied following the directional interpolation processing, as required.