Image Signal Processor Bit Depth Segmentation for HDR Detail Preservation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing techniques face challenges in preserving highlight and shadow details when reducing high-bit-depth image data to lower bit depths, often resulting in loss of dynamic range and requiring multiple exposures, which is inefficient and difficult to implement, especially with moving subjects.
Innovation Solution
An image signal processor with a high-bit-depth front-end and low-bit-depth back-end uses multiple transfer functions to convert high-bit-depth HDR image data into lower-bit-depth data, preserving details by blending based on local brightness levels, allowing for high-local-contrast image generation without multiple exposures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If multiple images at various exposure levels are captured and combined to preserve HDR details, then highlight and shadow details are preserved, but memory and processor usage increase and power consumption rises
Solution Approach 1:
The patent divides the high dynamic range into multiple segments, each handled by a dedicated data path with appropriate bit depth. A first data path processes highlights with higher bit depth while a second data path processes mid-tones and shadows with lower bit depth, eliminating the need for multiple image captures and reducing power consumption.
Solution Approach 2:
The patent transitions from temporal dimension (multiple images taken at different exposure times) to spatial dimension (multiple data paths processing different dynamic range segments simultaneously within a single image), thereby preserving HDR details without requiring multiple exposures.
2Loss of information
If multiple images at various exposure levels are captured and combined to preserve HDR details, then highlight and shadow details are preserved, but processing complexity and time increase
Solution Approach 1:
The patent segments the image processing into distinct data paths: a first data path for highlight regions with higher bit depth and a second data path for other regions with lower bit depth. This segmentation simplifies processing by avoiding the complex alignment and blending operations required by traditional multi-exposure HDR methods.
Solution Approach 2:
The patent changes the bit depth parameter dynamically across different spatial regions of the image. Higher bit depth is applied only to highlight regions where it is needed, while lower bit depth is used elsewhere, optimizing processing complexity while preserving essential HDR details.
3Productivity
If bit depth is reduced from high to low for standard processing, then processing efficiency increases, but dynamic range and detail information are lost
Solution Approach 1:
The patent applies different bit depth qualities to different regions of the image based on local requirements. Highlight regions maintain higher bit depth to preserve dynamic range information, while other regions use lower bit depth for processing efficiency, achieving both goals simultaneously.
Solution Approach 2:
The patent implements variable bit depth parameters across the image data, transitioning from a uniform low bit depth approach to a spatially varying bit depth scheme where high bit depth is maintained only where necessary for preserving dynamic range in highlight areas.
4Productivity
If a single exposure is used instead of multiple exposures, then processing time and complexity decrease, but it becomes difficult to capture both highlights and shadows
Solution Approach 1:
The patent resolves the single-exposure limitation by introducing a data path dimension rather than relying on temporal dimension. Multiple data paths with different bit depths process a single exposure simultaneously, enabling capture of both highlights and shadows without requiring multiple exposures or increasing processing time.
Data Source
AI summary
Devices and methods for obtaining image data that preserves highlight details and shadow details from image data from a high dynamic range (HDR) image sensor are provided. For example, an image signal processor may include a first data path of a relatively high bit depth, front-end image processing logic, and a second data path of a relatively low bit depth. The first data path may receive HDR image data from an image sensor. The front-end image processing logic may convert the HDR image data into lower-bit-depth image data, while preserving highlight details and/or shadow details of the HDR image, by using at least two transfer functions that preserve different subsets of the high dynamic range. The second data path may output this lower-bit-depth image data to other image processing logic for additional image processing.