1

许多函数scipy.ndimage接受一个可选mode=nearest|wrap|reflect|constant参数,该参数确定如何处理函数需要来自图像区域之外的一些数据(填充)的情况。填充由本机代码中的NI_ExtendLine()内部处理。

我不想在填充数据上运行 ndimage 函数,而是只想使用与 ndimage 使用的填充模式相同的填充模式来获取填充数据。

这是一个示例(仅适用于 mode=nearest,假设为 2d 图像):

"""
Get padded data.  Returns numpy array with shape (y1-y0, x1-x0, ...)
Any of x0, x1, y0, y1 may be outside of the image
"""
def get(img, y0, y1, x0, x1, mode="nearest"):
    out_img = numpy.zeros((y1-y0, x1-x0))
    for y in range(y0, y1):
        for x in range(x0, x1):
            yc = numpy.clip(y, 0, img.shape[0])
            xc = numpy.clip(x, 0, img.shape[1])
            out_img[y-y0, x-x0] = img[yc, xc]
    return out_img

这是正确的,但速度很,因为它一次迭代一个像素。

最好(最快、最清晰、最 Pythonic)的方法是什么?

4

1 回答 1

3
def get(img, y0, y1, x0, x1, mode="nearest"):
    xs, ys = numpy.mgrid[y0:y1, x0:x1]
    height, width = img.shape

    if mode == "nearest":
        xs = numpy.clip(xs, 0, height-1)
        ys = numpy.clip(ys, 0, width-1)

    elif mode == "wrap":
        xs = xs % height
        ys = ys % width

    elif mode == "reflect":
        maxh = height-1
        maxw = width-1

        # An unobvious way of performing reflecting modulo
        # You should comment this
        xs = numpy.absolute((xs + maxh) % (2 * maxh) - maxh)
        ys = numpy.absolute((ys + maxw) % (2 * maxw) - maxw)

    elif mode == "constant":
        output = numpy.empty((y1-y0, x1-x0))
        output.fill(0) # WHAT THE CONSTANT IS

        # LOADS of bounds checks and restrictions
        # You should comment this
        target_section = output[max(0, -y0):min(y1-y0, -y0+height), max(0, -x0):min(x1-x0, -x0+width)]
        new_fill = img[max(0, y0):min(height, y1), max(0, x0):min(width, x1)]

        # Crop both the sections, so that they're both the size of the smallest
        # Use more lines; I'm too lazy right now
        target_section[:new_fill.shape[0], :new_fill.shape[1]] = new_fill[:target_section.shape[0], :target_section.shape[1]]

        return output

    else:
        raise NotImplementedError("Unknown mode")

    return img[xs, ys]
于 2014-04-16T00:26:47.783 回答