Image Search Heat Map Generation for Subject Localization

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image processing systems face challenges in efficiently searching for specific subjects, such as faces, in images containing a large number of small subjects, leading to increased processing time and reduced search speed.

Innovation Solution

An image processing apparatus that extracts a feature vector from a query image and a feature vector map from the search image, performing arithmetic operations to generate a heat map indicating the likelihood of the subject's presence, allowing for rapid identification and localization of the search target.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If feature extraction and comparison are performed for each detected subject using conventional methods, then subject identification accuracy is improved, but processing time increases significantly when many small subjects are present

Engineering Contradiction:
Improvesubject identification accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent segments the image processing task into two distinct stages: (1) a coarse filtering stage using a lightweight model to quickly eliminate non-target regions, and (2) a precise identification stage using a sophisticated feature extractor only on the remaining candidate regions. This segmentation allows the system to maintain high accuracy while reducing overall processing time by avoiding exhaustive application of heavy computational methods across the entire image.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies preliminary action by performing initial subject detection and candidate region selection using a lightweight model before applying the more computationally intensive feature extraction and comparison operations. This preliminary filtering step reduces the number of subjects that require detailed analysis, thereby significantly reducing total processing time while maintaining identification accuracy for the final candidate set.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If a deep neural network with large number of parameters is used for high accuracy face identification, then identification accuracy is improved, but processing speed decreases

Engineering Contradiction:
Improveface identification accuracyVSAvoidprocessing speed
Core Design Contradiction:
Measurement precisionVSSpeed

Solution Approach 1:

The patent divides the processing workload by separating the deep neural network application to only candidate regions identified by a preliminary lightweight model. This segmentation ensures that the computationally expensive high-accuracy model is applied minimally, preserving processing speed while maintaining identification accuracy where it matters most.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies different processing qualities to different regions of the image: lightweight processing for most of the image and high-quality deep neural network processing only for candidate subject regions. This local quality approach optimizes the balance between accuracy and speed by concentrating computational resources where they are most needed.

Inventive Principle:
Principle #3Local quality

Data Source

PatentUS12197497B2Image processing apparatus for search of an image, image processing method and storage medium
Publication Date: 2025.01.14 CANON KK
  • US12197497B2 patent drawing
  • US12197497B2 patent drawing
  • US12197497B2 patent drawing

AI summary

An image processing apparatus is configured to acquire a query image including a subject that is a search target, and an image to be searched, extract a first feature vector that represents features of the search target included in the query image, extract a feature vector map that represents feature vectors of a subject at each position of the image to be searched, perform an arithmetic operation based on the first feature vector and the feature vector map to obtain a heat map that represents the likelihood that the search target is present.