Last Revision Date: 5/14/2015
This article describes the four kinds of stereo validation: texture, uniqueness, surface, and back and forth validation.
There are four kinds of validation used in our stereo vision systems
- Texture validation - indicates that the features matched between images do not have enough contrast or texture in them. Thresholding the edge strength inside the stereo mask
- Uniqueness validation - uniqueness validation was implemented before the additions of surface and back and forth validations. Since these validations are superior we do not recommend the usage of uniqueness validation.
- Surface validation - surface validation is unique to our stereo vision cameras.
- Back and forth - this validation verifies that the match chosen in the one image when the left image is the reference image is the same match as is chosen when the right image is the reference image. We recommend this validation be used, however the process of matching twice increases the resource cost of validation considerably. This validation is discussed in the Pascal Fua paper at Fua91.pdf