HDR Image Fusion Using Predicted Motion Vectors to Eliminate Ghosting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating High Dynamic Range (HDR) images struggle with ghost effects when capturing moving subjects across different exposure levels, particularly in areas with no information, leading to loss of detail and subject separation.
Innovation Solution
A method involving a processing unit that acquires multiple frames, calculates motion vectors, predicts motion detection matrices, and fuses frames based on these matrices to predict subject positions and eliminate ghost effects by considering pixel weights and exposure conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Illumination intensity
If multiple photographs are captured at different exposure levels and combined to produce HDR images, then the luminance range is improved, but ghost effects are produced around moving subjects
Solution Approach 1:
The patent applies preliminary action by predicting motion vectors for abnormal exposure frames based on normal exposure frames before the fusion process. Motion vectors for low-exposure and high-exposure frames are predicted using motion information from normal-exposure frames, allowing the system to pre-correct for moving subject positions before combining the frames, thereby preventing ghost effects while maintaining broad luminance range
Solution Approach 2:
The patent uses normal-exposure frames as an intermediary to transfer motion information to abnormal exposure frames. By calculating motion vectors from normal-exposure frames and applying them to low-exposure and high-exposure frames, the system creates a bridge that allows accurate motion compensation across all exposure levels, eliminating ghost effects while preserving the HDR luminance benefits
2Loss of information
If motion vectors are calculated from abnormal exposure frames, then motion information is obtained, but no reliable information is available in saturated or dark areas
Solution Approach 1:
The patent uses normal-exposure frames as an intermediary source to obtain reliable motion information. Since normal-exposure frames contain valid pixel information in all areas (unlike saturated or dark areas in abnormal exposure frames), motion vectors calculated from these frames provide accurate motion data that can be applied to correct the abnormal exposure frames without suffering from information loss in extreme exposure regions
3Illumination intensity
If frames are fused using traditional HDRM methods, then luminance compensation is achieved, but ghost effects cannot be eliminated when subjects move into exposure areas with no information
Solution Approach 1:
The patent applies preliminary action by pre-calculating motion vectors for abnormal exposure frames using motion information from normal exposure frames before the fusion process. This allows the system to predict where moving subjects will be in low-exposure and high-exposure frames, and use this information to guide the fusion process in selecting appropriate pixel values, thereby preventing ghost effects before they occur
Solution Approach 2:
The patent implements feedback by using motion vectors derived from normal-exposure frames to inform and adjust the fusion process for abnormal exposure frames. The motion information acts as feedback that guides the selection of pixel values during fusion, allowing the system to adaptively choose between normal and abnormal exposure frame pixels based on predicted subject positions, thereby eliminating ghost effects while maintaining luminance compensation
Data Source
AI summary
A method for generating HDR (High Dynamic Range) images, performed by a processing unit, is introduced to at least contain: acquiring a frame 0 and a frame 1; calculating a first MV (Motion Vector) between the frame 0 and the frame 1; acquiring a frame 2; predicting a second MV between the frame 0 and the frame 2 according to the first MV, a time interval between shooting moments for the frames 0 and 1 and a time interval between shooting moments for the frames 0 and 2; generating a first MD (Motion Detection) matrix comprising a plurality of first MD flags according to the second MV; and fusing the frame 0 with the frame 2 according to the first MD flags.