Sub-sampled Integral Images for Efficient Face Detection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional face detection algorithms are inefficient when run on GPUs due to high computational requirements and decreased localization of processed data as classifier scale or size increases, leading to reduced processing efficiency.

Innovation Solution

A computer-implemented method that calculates the integral image of an image and sub-samples it to generate sub-sampled integral images, applying classifiers in stages to these regions to generate classification data, which is used to determine if a face is present, thereby reducing the number of integral image calculations and increasing localized data processing efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If classifiers of different scales are applied to the integral image to detect faces, then face detection coverage is improved, but localization of processed data decreases as classifier scale increases

Engineering Contradiction:

Improveface detection coverageVSAvoidlocalization of processed data

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent divides the integral image into multiple smaller images by removing every nth row and column, creating a segmented structure that allows classifiers to process localized data efficiently while maintaining detection coverage across different scales

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent transforms the problem from processing the entire integral image at once to processing multiple smaller images, adding a dimensional aspect of image decomposition that enables both comprehensive detection and localized processing

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

2Measurement precision

If multiple integral images are calculated for downscaled images, then face detection accuracy is improved, but computational intensity increases

Engineering Contradiction:

Improveface detection accuracyVSAvoidcomputational intensity

Core Design Contradiction:

Measurement precisionVSPower

Solution Approach 1:

The patent performs preliminary action by calculating the integral image only once for the original image, then reuses this single integral image for detecting faces at multiple scales through sub-sampling, eliminating the need to calculate multiple integral images

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The single integral image serves multiple functions by being sub-sampled to create multiple smaller images for different scale detections, making the integral image calculation universally applicable across all detection scales

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS8442327B2Application of classifiers to sub-sampled integral images for detecting faces in images

Publication Date: 2013.05.14 NVIDIA CORP

AI summary

A method for more efficiently detecting faces in images is disclosed. The integral image of an image may be calculated. The integral image may be sub-sampled to generate one or more sub-sampled integral images. A plurality of classifiers may be applied in one or more stages to regions of each sub-sampled integral image, where the application of the classifiers may produce classification data. The classification data may be used to determine if a face is associated with any of the regions of each sub-sampled integral image. The face determination results may be used to modify the original image such that, when rendered, the image is displayed with a graphical object identifying the face in the image. Accordingly, face detection processing efficiency may be increased by reducing the number of integral image calculations and processing localized data through application of classifiers to sub-sampled integral images.