Photometric Stereo Surface Reconstruction Under Unknown Illumination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing photometric stereo methods are inadequate for generating accurate raised relief maps of objects with complex illumination conditions and unknown illumination directions, particularly in biological analysis, as they require prior information about illumination sources and surface normals which is often unavailable.
Innovation Solution
A method that adapts model components to reproduce surface properties and illumination conditions without knowing the position or direction of illumination sources, using structuring component matrices to represent albedo values and intensity coefficients uniformly across pixels, enabling the generation of raised relief maps from multiple images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional photometric stereo algorithms are used, then raised relief maps can be generated under simple illumination conditions, but the algorithms fail when illumination sources are not point sources or when illumination directions are unknown
Solution Approach 1:
The patent changes the parameters of the photometric stereo model by introducing unknown illumination directions and non-point light sources as variables to be optimized, rather than treating them as fixed known parameters. This allows the algorithm to adapt to complex illumination conditions while maintaining reconstruction accuracy through iterative optimization of these parameters alongside surface normals and albedo values.
2Ease of operation
If approximations are made to apply existing algorithms, then processing can proceed with unknown illumination information, but the representation of the object surface becomes incorrect
Solution Approach 1:
The patent implements a feedback mechanism through iterative optimization where the algorithm continuously refines its estimates of illumination directions, light source positions, surface normals, and albedo values. Each iteration uses the reconstructed surface information to improve illumination parameter estimation, which in turn improves the accuracy of the raised relief map, creating a self-correcting process that eliminates the need for initial approximations.
3Productivity
If illumination source positions and directions are assumed known, then the algorithm can operate, but it cannot handle biological samples with complex or unknown illumination conditions
Solution Approach 1:
The patent creates a universal photometric stereo algorithm that can handle multiple types of illumination conditions (point sources, extended sources, unknown directions, known directions) within a single unified framework. By treating illumination parameters as optimizable variables rather than fixed inputs, the algorithm becomes universally applicable to both simple and complex illumination scenarios, including biological sample analysis with unknown illumination conditions.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method for obtaining a model of an object surface from a plurality of images of said object, wherein the object is illuminated with one or more illumination sources, the method comprising: • - obtaining a set of images each comprising an array of pixels and representing similar views of an object, wherein the similar views are obtained under different illumination conditions; • - defining a model function that expresses the known pixel intensity values of the images in terms of the following unknown model components: • - a first model component (A) representing the albedo of the object at each pixel location and being the same for the plurality of images and being an albedo value; • - a second model component (L) representing an intensity of illumination for each image and being an illumination source intensity value being the same for all pixels of each image; • - a third model component (V) representing a specific illumination direction and being different for each image and being an illumination vector and being the same for all pixels of each image; • - a fourth model component (N) representing surface normal directions of the object surface at each pixel position and being the same for all images and being a normal vector; • - performing one or more sequences of minimization operations to minimize a difference function between the pixel values obtained from the set of images and pixel values calculated using said model function, each minimization operation being performed by allowing one of said model components (A, L, V, N) to vary while the others remain unchanged; • - outputting the fourth model component (N) as said model of the object surface.