Face Detection Using Skin Tone Variability Rejection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing face detection techniques are memory-intensive and fail to achieve processing speeds of 30 frames per second, especially when scene changes occur, limiting their effectiveness in applications like video indexing and content mining.

Innovation Solution

A fast face detection method that increases the rejection rate of non-face rectangles using skin tone probability and variability values, reducing the complexity of cascade levels and employing skin tone and variability parameters for rapid rejection, allowing for efficient detection and tracking of human faces.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Speed

If traditional face detection techniques are used, then detection accuracy is maintained, but processing speed is slow and memory consumption is high

Engineering Contradiction:
Improveprocessing speedVSAvoidmemory demand
Core Design Contradiction:
SpeedVSDevice complexity

Solution Approach 1:

The patent segments the face detection process into multiple cascade levels (first level, second level, etc.), where each level performs progressively more complex analysis. The first level uses simple skin tone probability and variability checks to quickly reject non-face regions, while subsequent levels perform more detailed verification only on regions that pass the initial filter. This segmentation allows the system to achieve high processing speed at early stages while maintaining detection accuracy through progressive refinement.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies partial action by performing comprehensive face detection analysis only on regions that pass preliminary filtering criteria. Instead of analyzing every region in the image with full complexity, the system performs skin tone-based rejection on all regions first, then applies more computationally intensive verification only to candidate regions. This reduces overall computational load and memory demand while maintaining detection accuracy for actual face regions.

Inventive Principle:
Principle #16Partial or excessive action

2Productivity

If traditional face detection techniques are used, then detection accuracy is maintained, but processing speed cannot reach 30 frames per second

Engineering Contradiction:
Improveframes per secondVSAvoidprocessing time
Core Design Contradiction:
ProductivityVSLoss of time

Solution Approach 1:

The patent performs preliminary skin tone probability and variability analysis on all image regions before conducting detailed face verification. By pre-filtering regions based on skin tone characteristics at the first cascade level, the system eliminates non-face regions early in the processing pipeline, reducing the time required for subsequent verification steps. This preliminary action enables the system to process images at 30 frames per second by minimizing the computational workload for each frame.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent implements skipping by rapidly rejecting non-face regions through skin tone-based filtering without performing comprehensive verification. The first cascade level quickly skips over regions that fail skin tone probability or variability thresholds, advancing directly to verification only for candidate face regions. This rushing through of non-relevant regions significantly reduces processing time per frame, enabling real-time detection at 30 frames per second.

Inventive Principle:
Principle #21Skipping (Rushing through)

Data Source

PatentUS9053354B2Fast face detection technique
Publication Date: 2015.06.09 INTEL CORP
  • US9053354B2 patent drawing
  • US9053354B2 patent drawing
  • US9053354B2 patent drawing

AI summary

A processor includes a face recognition block and a face detection block. The face detection block includes a scan block and control logic. The scan block may divide an image into a plurality of rectangles, determine whether at least some pixel values in each of the plurality of rectangles is within an allowable skin tone range, reject one or more of the plurality of rectangles, which do not contain pixels having a value within the allowable skin tone range, and mark remaining of the plurality of rectangles as component rectangles of a human face. Further, the scan block is to determine variability of the component rectangles, compare the variability of the component rectangles with a variability threshold, reject one or more of the component rectangles whose variability is less than the variability threshold, and retain remaining component rectangles as features of the human face.