Homomorphic Image Processing for Night Vision Contrast Enhancement

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

High-dynamic, high-resolution digital video imaging systems, especially in low-light conditions, face challenges such as noise amplification, degradation of wide luminous dynamics, appearance of artifacts, and incompatibility with real-time processing at high frame rates, leading to decreased image quality and loss of scene aspect.

Innovation Solution

A homomorphic filtering method that processes images by calculating the logarithm of the initial image, applying a low-frequency Gaussian filter, determining intensity values, and performing a linear combination to enhance contrast without increasing noise, suitable for real-time operation and dynamic adaptation between sensor and display.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional contrast enhancement methods are applied to low-light digital video images, then image contrast is improved, but digital noise is strongly increased

Engineering Contradiction:
Improveimage contrastVSAvoiddigital noise
Core Design Contradiction:
Measurement precisionVSObject-generated harmful factors

Solution Approach 1:

The image processing method segments the image into different frequency components using a low-pass filter. The low-frequency components (representing overall illumination) are separated from high-frequency components (representing fine details and noise). This segmentation allows selective enhancement of contrast in low-frequency regions without amplifying noise in high-frequency regions.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies different processing characteristics to different frequency regions. The low-pass filtering operation applies a smoothing kernel that preserves low-frequency information while attenuating high-frequency noise. The contrast enhancement is then applied selectively to the filtered low-frequency components, creating local quality differentiation between frequency domains.

Inventive Principle:
Principle #3Local quality

2Measurement precision

If conventional contrast enhancement methods are applied, then low gray level contrasts are enhanced, but saturation of overexposed areas occurs

Engineering Contradiction:
Improvelow gray level contrastVSAvoidoverexposed area integrity
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

By segmenting the image into low-frequency and high-frequency components, the method processes different regions with different characteristics. The low-pass filter separates the overall illumination pattern from local variations, allowing contrast enhancement to be applied primarily to the low-frequency component without affecting the dynamic range of overexposed high-frequency regions.

Inventive Principle:
Principle #1Segmentation

3Measurement precision

If conventional image processing methods are applied, then contrast is enhanced, but processing compatibility with real-time operation at high frame rates is lost

Engineering Contradiction:
Improveimage contrastVSAvoidreal-time processing capability
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The method segments the computationally intensive contrast enhancement task into a two-stage process: first applying a low-pass filter to obtain low-frequency components, then enhancing contrast on this simplified representation. This segmentation reduces the overall computational burden compared to applying complex contrast enhancement to the full-resolution image.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies partial action by enhancing contrast only on the low-frequency component obtained through low-pass filtering, rather than processing the entire image at full resolution. This partial processing approach significantly reduces computational requirements while maintaining perceptual quality, enabling real-time operation at high frame rates.

Inventive Principle:
Principle #16Partial or excessive action

4Measurement precision

If conventional contrast enhancement methods are applied, then local contrasts are improved, but resolution of the initial image decreases

Engineering Contradiction:
Improvelocal contrastVSAvoidimage resolution
Core Design Contradiction:
Measurement precisionVSManufacturing precision

Solution Approach 1:

The patent segments the image processing into frequency domains using low-pass filtering. The contrast enhancement is applied to the low-frequency component, while the high-frequency component (containing fine detail and resolution information) is preserved. This segmentation allows local contrast improvement without degrading the resolution carried by high-frequency components.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

After separate processing of the low-frequency component (contrast enhancement via low-pass filtering), the method merges the processed low-frequency component back with the original high-frequency component. This merging operation combines the enhanced local contrast from the low-frequency part with the fine detail and resolution information from the high-frequency part, achieving both goals simultaneously.

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentEP2887307B1Image-processing method, in particular for images from night-vision systems and associated system
Publication Date: 2016.11.30 THALES SA
  • EP2887307B1 patent drawingFigure 1
  • EP2887307B1 patent drawingFigure 2
  • EP2887307B1 patent drawing

AI summary

The general field of the invention is that of methods for processing an initial digital image IIN(p) composed of pixels (p). The method according to the invention comprises the following successive steps: Step 1: Calculation of the logarithm of the initial image to obtain a first intermediate image denoted Log(IIN(p)); Step 2: Filtering of said first intermediate image by a low-frequency filter to obtain a second intermediate image denoted BF(p); Step 3: Determination of the minimum intensity value and the maximum intensity value in the low-frequency image; Step 4: Calculation of a third intermediate image denoted Log(IOUT(p)) by the following linear combination: LogIOUTp=Log(IINp-BFp.K⁢1.⁢BFp+K⁢2+K⁢3.⁢BFp+K⁢4 K1, K2, K3 and K4 being constants; Step 5: Calculation of the final image IOUT(p) by applying the exponential function to the third intermediate image.