Depth Estimation via Weighted Similarity and Transition Penalty
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current algorithms for estimating depth information values in digital images are either computationally complex and provide high fidelity or are simple and suitable for real-time processing but offer limited fidelity, making them unsuitable for high-fidelity real-time processing on mobile devices.
Innovation Solution
An image processing apparatus and method that calculates a similarity measure for a current fragment based on a depth information value candidate, compares it with a previously selected value, applies a weighting function to penalize transitions in depth information values, and determines the optimal depth information value for the current fragment, thereby achieving high fidelity in a computationally efficient manner.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sophisticated cost functions with min-convolution and transition cost functions are used for depth estimation, then measurement precision is improved, but device complexity increases making real-time processing impossible
Solution Approach 1:
The patent extracts and removes the computationally expensive min-convolution operation from the cost function. Instead of using the full dynamic programming approach with min-convolution to enforce depth continuity, the invention uses a simplified cost function that only includes local matching costs, thereby reducing complexity while maintaining acceptable depth estimation quality
Solution Approach 2:
The patent changes the parameter representation by using a penalty term based on depth discontinuities rather than full min-convolution. The cost function is modified to include a penalty term that penalizes large depth changes between neighboring pixels, achieving depth continuity enforcement with significantly reduced computational complexity
2Productivity
If simple local fragment matching algorithms are used for depth estimation, then productivity is improved enabling real-time processing, but measurement precision deteriorates
Solution Approach 1:
The patent applies local quality by making the cost function adaptive to local image characteristics. The penalty term for depth discontinuities is applied selectively based on local matching quality, allowing the algorithm to maintain high precision in difficult regions while keeping processing fast in simple regions
Solution Approach 2:
The patent introduces feedback mechanisms where the depth estimation process uses information from previously processed fragments to guide current fragment processing. The penalty term uses depth information from neighboring fragments to enforce continuity, creating a feedback loop that improves precision without significantly increasing processing complexity
Data Source
AI summary
The present disclosure can provide an image processing apparatus. In some embodiments, the image processing apparatus in accordance with the disclosure comprises memory configured to store a previously selected depth information value (dprev), (dprev; dprevi) having been selected for a previously processed fragment and signal processing logic (103). The signal processing logic is configured to calculate a similarity measure for the current fragment based on a depth information value candidate defining a reference fragment candidate of a digital reference image, compare the depth information value candidate with the previously selected depth information value (dprev), apply a weighting function to the calculated similarity measure to obtain a weighted similarity measure for the current fragment based on the comparison, and determine whether to select the depth information value candidate as the depth information value (dbest) for the current fragment based on the weighted similarity measure of the depth information value candidate.