Validity criteria

Pandora 2D utilizes seven criteria to either invalidate pixels that cannot be computed or identify specific pixels. During execution, these criteria are represented as 8-bit encoded flags and stored as a criteria data variable named criteria within the cost volumes. This data variable shares the same dimensions as the cost volumes: row, col, disp_row, and disp_col.

This approach enables the assignment of a criterion to each disparity pair for every pixel in the image, ensuring a clear representation of the underlying process.

Note

The criteria are cumulative, meaning multiple criteria can be applied to a single pixel—except at the image boundaries.

For ease of use, criterion values are managed as an enumeration (enum), allowing operations to be performed via their names rather than directly manipulating numerical values.

Ultimately, the criteria dataset serves as the foundation for constructing the validity dataset, which provides information regarding the reliability of each pixel in the image.

Below is a summary of each criterion (clic on criterion’s name to jump to its detailed description).

Table 1 Criteria summary

Criterion	Value in criteria data var	Description
P2D_LEFT_BORDER	2⁰ = 1	border of left image according to window size
P2D_LEFT_NODATA	2¹ = 2	nodata in left mask
P2D_RIGHT_NODATA	2² = 4	nodata in right mask
P2D_RIGHT_DISPARITY_OUTSIDE	2³ = 8	disparity is out the right image
P2D_INVALID_MASK_LEFT	2⁴ = 16	invalidated by validity mask of left image
P2D_INVALID_MASK_RIGHT	2⁵ = 32	invalidated by validity mask of right image
P2D_PEAK_ON_EDGE	2⁶ = 64	The correlation peak is at the edge of disparity range. The calculations stopped at the pixellic stage
P2D_INVALID_INIT_DISPARITY	2⁷ = 128	The initial disparity is invalid

Detailed Criteria’s explanations

P2D_LEFT_BORDER

This criterion excludes edge pixels where the matching cost window would extend beyond the image boundaries. Since this window is centered on the pixel and has an odd size, the offset considered follows this formula:

Where window_size is an input parameter of matching cost.

Note

This criterion results from left image and is not dependent upon disparity. As a result, this criterion will be assigned to all disparities associated to this pixel in the criteria data array.

Warning

When this criterion is raised for a pixel, no other criterion is raised for this pixel.

P2D_LEFT_NODATA

This criterion is raised when NO DATA is found in the matching cost window. To identify the affected pixels, a binary dilation, equivalent in size to the matching cost window, is applied to the pixels corresponding to NO DATA in the left image mask.

Note

This criterion results from left image and is not dependent upon disparity. As a result, this criterion will be assigned to all disparities associated to this pixel in the criteria data array.

P2D_RIGHT_NODATA

This criterion is raised when a pixel within the matching cost window of the explored pixel in the right image is marked as NO DATA in the right mask.

Note

The criterion is raised for sub-pixel disparities if a no-data value from the input mask was used to interpolate the corresponding point.

P2D_RIGHT_DISPARITY_OUTSIDE

This criterion is raised when the matching cost window goes outside the right image.

P2D_INVALID_MASK_LEFT

This criterion is raised when the pixel is marked as INVALID in the left mask.

Note

This criterion results from left image and is not dependent upon disparity. As a result, this criterion will be assigned to all disparities associated to this pixel in the criteria data array.

P2D_INVALID_MASK_RIGHT

This criterion is raised when the explored pixel in the right image is marked as INVALID in the right mask.

Note

Ignore invalid pixels from the input mask used in interpolation to raise the criterion.

P2D_PEAK_ON_EDGE

The best similarity measure selected by the Winner Takes All algorithm is an extremum. If the selected extremum is on the edge of the cost surface, we cannot determine whether it is a true extremum or merely point of a trend cut off by the chosen disparity range.

In this case, the P2D_PEAK_ON_EDGE criterion is raised.

P2D_INVALID_INIT_DISPARITY

This criterion is raised when the initial disparity of the pixel is invalid.

This situation can arise when providing a variable initial disparity (see Inputs) that contains points with a value of no data in the metadata of the initial disparity grid. In particular, when using disparity maps previously calculated by pandora2d as initial disparity grids, the invalid pixels in these initial disparity grids correspond to pixels that could not be calculated by pandora2d during the previous execution.

Note

This criterion results from invalid points in initial disparity grids and it does not depend on the disparity values of the current execution. As a result, this criterion will be assigned to all disparities associated to this pixel in the criteria data array.

Validity dataset

From the criteria data array, the validity dataset is generated to summarize the information and store it. It consists of multiple data variables, one for each criterion, along with the validity data variable. These variables are saved as separate bands, named after their corresponding criterion, in a TIFF file named validity.tif, located in the output/disparity_map directory.

Criterion bands

A separate band is created for each criterion. If a criterion is present in the cost surface mask of a given pixel in the image, the corresponding pixel in the band is set to 1; otherwise, it is set to 0.

Validity

If at least one requested disparity can be computed in the cost surface mask of a pixel, this pixel is considered as PARTIALLY_VALID, and encoded by a 0 in the partial_validity_mask. If all the required disparity pairs can not be computed: in this case, the pixel is considered as INVALID and encoded as 1.

In the validity_mask band, VALID pixels are encoded as 0 when all requested disparities can be computed, otherwise INVALID pixels are encoded as 1.