crop.py

import numpy as np

def safe_crop(arr, tblr, fill=None):
    n,s,w,e = tblr
    shape = np.asarray(arr.shape)
    shape[:2] = s - n, e - w
    no, so, wo, eo = 0, shape[0], 0, shape[1]
    if n < 0:
        no += -n
        n = 0
    if w < 0:
        wo += -w
        w = 0
    if s >= arr.shape[0]:
        so -= s - arr.shape[0]
        s = arr.shape[0]
    if e >= arr.shape[1]:
        eo -= e - arr.shape[1]
        e = arr.shape[1]
    cropped = arr[n:s,w:e]
    if fill is None:
        return cropped
    out = np.empty(shape, dtype=arr.dtype)
    out.fill(fill)
    try:
        out[no:so,wo:eo] = cropped
    except ValueError:
        # this happens when there is no overlap
        pass
    return out

def place_inside(img, xy, output_shape, fill=0):
    ix,iy = xy
    ih,iw,ic = img.shape
    oh,ow = output_shape
    out = np.empty((oh,ow,ic), dtype=img.dtype)
    out.fill(fill)
    ox = 0
    oy = 0
    if ih < oh and iy == 0:
        oy = oh - ih
    if iw < ow and ix == 0:
        ox = ow - iw
    out[oy:oy+ih,ox:ox+iw] = img
    return out

def inner_square_crop(img):
    size = np.asarray(img.shape[:2])
    min_side = min(size)
    corner = (size - min_side) // 2
    return img[corner[0]:corner[0]+min_side, corner[1]:corner[1]+min_side]

def outer_square_crop(img, fill=0):
    size = np.asarray(img.shape[:2])
    max_side = max(size)
    output_shape = np.copy(img.shape)
    output_shape[:2] = max_side
    out = np.empty(output_shape, dtype=img.dtype)
    out.fill(fill)
    corner = (max_side - size) // 2
    out[corner[0]:corner[0]+size[0], corner[1]:corner[1]+size[1]] = img
    return out