Multi-viewpoint Image Encoding Parallax Compensation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multi-viewpoint image encoding techniques face inefficiencies in parallax compensation, particularly when dealing with decimal pixel positions, leading to increased computation costs and prediction errors due to the need for high precision and large amounts of encoded data.
Innovation Solution
The method involves determining corresponding points on a target image based on distances from the camera to the imaged object, using parallax vectors and target predictive vectors to perform inter-camera image prediction, allowing for accurate parallax compensation at decimal pixel positions while minimizing the number of encoded parallax data items.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high precision parallax compensation is performed at decimal pixel positions, then prediction accuracy is improved, but computation cost increases
Solution Approach 1:
The patent segments the image into multiple blocks and performs parallax compensation independently for each block. By dividing the image processing into smaller units, the computation is distributed and managed more efficiently, allowing high-precision decimal pixel position compensation without overwhelming computational burden on the entire image at once.
Solution Approach 2:
The patent applies parallax compensation selectively to specific blocks rather than uniformly to the entire image. By identifying and processing only those blocks that require parallax compensation (partial action), the computation cost is reduced while maintaining prediction accuracy where needed. This avoids excessive computation in areas where simple interpolation suffices.
2Measurement precision
If parallax data is encoded for each target frame, then prediction accuracy is improved, but the amount of encoded data increases
Solution Approach 1:
The patent merges parallax data from multiple target frames and combines it with reference image data. By integrating parallax information across different frames and combining with reference images, the system achieves accurate prediction while encoding fewer redundant parallax parameters. This merging reduces the total quantity of encoded data compared to encoding separate parallax data for each frame.
Solution Approach 2:
The patent creates a universal reference image that serves multiple target frames simultaneously. This reference image contains parallax compensation information that can be applied to predict multiple different target frames, making the encoded data multi-functional. Instead of encoding separate parallax data for each target frame, one reference image with parallax data serves universal prediction needs, significantly reducing the amount of encoded data.
Data Source
AI summary
An image encoding method includes determining a corresponding point on a target image for encoding, which corresponds to each pixel on a reference image, based on the distance from a camera used for obtaining the reference image to an imaged object, and a positional relationship between cameras; computing a parallax vector from the position of the pixel to the corresponding point in the pixel space; computing a target predictive vector having the same starting point as the parallax vector and components obtained by rounding off the components of the parallax vector; computing a target reference vector having the same starting point as the parallax vector and the same size and direction as a differential vector between the target predictive vector and the parallax vector; and setting a predicted pixel value of a pixel on the target encoding image, which is indicated by the target predictive vector, to a value of a pixel on the reference image, which is indicated by the target reference vector.