Depth Map Shaping via Bilateral Filtering for Low Resolution Output

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for shaping depth maps do not effectively accommodate low resolution depth information to the correct contours of an object without up-sampling, leading to inaccuracies in representation.

Innovation Solution

An image processing apparatus that acquires depth distribution information with a first resolution and image data with a higher second resolution, then shapes the map data by resizing the image data to match the first resolution, using bilateral filter processing to ensure accurate contour representation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If up-sampling is performed to generate high resolution depth map, then the resolution of output map data is improved, but power consumption increases and shaping accuracy at low resolution deteriorates

Engineering Contradiction:
Improvedepth map resolutionVSAvoidpower consumption
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent changes the resolution parameter of the depth map to match the lower resolution requirement, avoiding unnecessary up-sampling operations. This parameter adjustment resolves the contradiction by accepting lower output resolution to reduce power consumption while maintaining adequate shaping accuracy through the bilateral filtering process applied at the target resolution

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If up-sampling is performed to generate high resolution depth map, then the resolution of output map data is improved, but shaping accuracy at low resolution deteriorates

Engineering Contradiction:
Improvedepth map resolutionVSAvoidcontour shaping accuracy
Core Design Contradiction:
Measurement precisionVSManufacturing precision

Solution Approach 1:

Instead of up-sampling low resolution depth data to high resolution and then shaping, the patent inverts the approach by shaping the depth map at the lower target resolution directly. This inversion prevents the introduction of errors that would occur during up-sampling and subsequent down-sampling, thereby maintaining better contour shaping accuracy at the intended output resolution

Inventive Principle:
Principle #13The other way round (Inversion)

3Use of energy by moving object

If image data is resized to match low resolution map data, then power consumption is reduced and low resolution shaping accuracy is improved, but high resolution detail information is lost

Engineering Contradiction:
Improvepower consumptionVSAvoidimage detail information
Core Design Contradiction:
Use of energy by moving objectVSLoss of information

Solution Approach 1:

The patent extracts only the necessary depth information at the target low resolution from the high resolution image data, rather than processing or retaining all high resolution details. The bilateral filtering operation extracts depth contours while discarding high frequency detail information that would be lost anyway in the low resolution output, thus minimizing information loss relevant to the application

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentUS10204400B2Image processing apparatus, imaging apparatus, image processing method, and recording medium
Publication Date: 2019.02.12 CANON KK
  • US10204400B2 patent drawing
  • US10204400B2 patent drawing
  • US10204400B2 patent drawing

AI summary

An imaging unit images an object through an imaging optical system so as to acquire image data. A depth map acquiring unit acquires information relating to a depth distribution of an object as depth map data. The resolution of depth map data is relatively lower than the resolution of image data which has been imaged. A depth map shaping unit references image data of an object so as to conform to the resolution of a depth map, when it performs shaping of the depth map based on image data of the object.