公开(公告)号：US20140074852A1

公开(公告)日：2014-03-13

申请号：US14076890

申请日：2013-11-11

Applicant: Microsoft Corporation

Inventor： Avneesh Sud ,  Rajeev Prasad ,  Ayman Malek Abdel Hamid Kaheel ,  Pragyana Mishra ,  Sumit Amar ,  Kancheng Cao

IPC: G06F17/30

CPC classification number: G06F17/30256 ,  G06F17/30277 ,  G06F17/30619 ,  G06F17/30651 ,  G06F17/30864

Abstract: Systems, methods, and computer-readable storage media for web-scale visual search capable of using a combination of visual input modalities are provided. An edgel index is created that includes shape-descriptors, including edgel-based representations, that correspond to each of a plurality of images. Each edgel-based representation includes pixels that depicts edges or boundary contours of an image and is created, at least in part, by segmenting the image into a plurality of image segments and performing a multi-phase contour detection on each segment. Upon receiving a search query having a visual query input, the visual query input is converted into shape-descriptors, including an edgel-based representation, and the shape-descriptors, including the edgel-based representation, of each of the plurality of images is compared with the shape-descriptors, including the edgel-based representation, of the visual query input to identify at least one image of the plurality of images that matches the visual query input.

Abstract translation: 提供了能够使用视觉输入模式的组合的用于web尺度视觉搜索的系统，方法和计算机可读存储介质。 创建包含与多个图像中的每一个相对应的形状描述符（包括基于灰色的表示）的暗角指数。 每个基于灰色的表示包括描绘图像的边缘或边界轮廓的像素，并且至少部分地通过将图像分割成多个图像片段并且在每个片段上执行多相轮廓检测来创建。 在接收到具有视觉查询输入的搜索查询时，视觉查询输入被转换成形状描述符，包括基于边缘的表示，并且包括基于灰色的表示的形状描述符，包括多个图像中的每一个的形状描述符是 与形状描述符（包括基于边缘的表示）相比较，用于识别与视觉查询输入匹配的多个图像中的至少一个图像的视觉查询输入。

公开(公告)号：US10509963B2

公开(公告)日：2019-12-17

申请号：US13722085

申请日：2012-12-20

Applicant: MICROSOFT CORPORATION

Inventor： Ayman Malek Abdel Hamid Kaheel ,  Padma Priya Gaggara ,  Prakash Asirvatham Arul ,  Mohammad Adil Hafeez ,  Dhananjay Dilip Kulkarni ,  Kancheng Cao

IPC: G06K9/00 ,  G06F16/951 ,  G06F16/9535

Abstract: Systems, methods, and computer storage media for discovering authoritative images of people entities are provided. Selections of person entities are received. Authoritative URLs and authoritative images for the person entities are identified. Once the authoritative images are identified, features are extracted. Queries for the person entities are identified by mining search engine logs. The queries and features can be utilized to construct candidate queries to identify and retrieve candidate image URLs. Candidate features are extracted for each candidate image associated with the candidate image URLs. Training data may be utilized to train a classifier that can be run on each candidate image. Each candidate image can then be tagged with an entity ID tag. Images with the entity ID tag can be ranked higher in search engine results page than images without the entity ID tag.