Interpolation of dead or missing pixels

[tifuchs]

Currently, missing pixels are excluded from the integration and their value approximated by the average value of the other 'good' pixels.

This seems to be a better approximation than effectively setting the value of the pixels to zero.

I asked AI for a better, but still simple solution. One good idea would be to perform a local interpolation:

@njit(inline='always')
def interpolate_pixel(image, mask, i, j):
    h, w = image.shape

    # 4-connected neighbors
    s = 0.0
    c = 0

    if i > 0 and not mask[i-1, j]:
        s += image[i-1, j]; c += 1
    if i < h-1 and not mask[i+1, j]:
        s += image[i+1, j]; c += 1
    if j > 0 and not mask[i, j-1]:
        s += image[i, j-1]; c += 1
    if j < w-1 and not mask[i, j+1]:
        s += image[i, j+1]; c += 1

    if c > 0:
        return s / c

    # fallback: 8-connected neighbors
    s = 0.0
    c = 0
    for di in (-1, 0, 1):
        for dj in (-1, 0, 1):
            if di == 0 and dj == 0:
                continue
            ni = i + di
            nj = j + dj
            if 0 <= ni < h and 0 <= nj < w:
                if not mask[ni, nj]:
                    s += image[ni, nj]
                    c += 1

    if c > 0:
        return s / c

    return np.nan

This would be a better fix for single or few clustered missing pixels.

However, the elephant in the room: Detector gaps, are not solved by this.