DYNAMIC IMAGE SEARCH BASED ON USER-SELECTED OBJECTS

Smart overview of the Invention

The patent application discusses a novel approach to image searching, focusing on user-selected objects within an image rather than the entire image. Traditional search engines typically process the whole image to find similar ones, but this method allows users to pinpoint specific objects of interest. The approach involves identifying multiple objects in an input image, selecting desired ones, and creating a query image from those selections. This query image is then used to find related images that match the user's specific interests.

Background

Conventional image search methods have limitations when users are interested in particular parts of an image. Users often resort to manually cropping images to focus on specific objects, but this can be challenging and may not yield accurate results. Additionally, objects may be partially obscured or require different spatial arrangements, complicating manual modifications. The proposed system addresses these challenges by allowing users to select and rearrange objects within an image to create a more targeted query.

Summary of Invention

The invention provides a method implemented on computing devices where an input image is analyzed to detect multiple objects. Users can select two or more of these objects to form a query image, which is then used to perform a search for similar images. The system can also adjust the positioning of selected objects within the query image, enabling searches that reflect potential changes in object relationships, such as proximity alterations.

Machine Learning Integration

Machine learning techniques are integral to the system's functionality, particularly in object detection and image processing tasks. Various models like neural networks are employed for operations such as object segmentation and embedding mapping. These models undergo training phases using both supervised and unsupervised learning methods to refine their accuracy in recognizing and processing selected objects from input images.

Definitions and Terminology

Key terms include "input image," referring to the initial image used for search; "object," denoting entities within images; "object mask," which involves modifying an image to isolate selected objects; and "composite image," created from object masks used as query images for searching similar visuals. The application extends the definition of "image" to encompass both still pictures and video content.