从逐块处理函数中调用过滤器函数 [英] Calling filter functions from within a block-by-block processing function

问题描述

我有一些内存密集型图像过滤器，我想在大型图像/数组上逐块调用（因为它们为整个数组计算过滤器，在尝试计算整个数组时它们会耗尽内存）。 p>

I have some memory intensive image filters that I want to call block by block on large images/arrays (because they compute the filter for an entire array, they run out of memory when trying to compute the whole array).

def block_process(Ic, blocksize):
    B = numpy.empty(Ic.shape)

    colstart = 0
    while colstart < Ic.shape[1]:
        BlockWidth = blocksize
        if (colstart + blocksize) > Ic.shape[1]:
            BlockWidth = Ic.shape[1] - colstart
        rowstart = 0
        while rowstart < Ic.shape[0]:
            BlockHeight = blocksize
            if (rowstart + blocksize) > Ic.shape[0]:
                BlockHeight = Ic.shape[0] - rowstart

            B[colstart:colstart+BlockWidth, rowstart:rowstart+BlockHeight] = filter1(params) # One of many available filters

            rowstart += BlockHeight
        colstart += BlockWidth
    return B # The complete filtered array

我的过滤器在其他函数中计算，即 def filter1（A，filtsize）， def filter2（A，filtsize，otherparam），它有一个 A 参数（输入数组，由块函数给出），以及其他参数如作为过滤器大小。某些过滤器的参数比其他过滤器多。他们返回过滤后的数组。

My filters are computed in other functions i.e. def filter1(A, filtsize), def filter2(A, filtsize, otherparam), which have an A parameter (the input array, given by the block function), and other parameters such as filter size. Some filters have more parameters than others. They return the filtered array.

两个问题

1. 如何通过block_process函数调用我的一个过滤函数？我不想将块处理代码复制到每个函数中。换句话说，有没有办法指定调用哪个过滤器（以及使用什么参数）作为 block_process（）调用的参数？


2. 有更好的编码方法吗？

1. How do I go about calling one of my filter functions through the block_process function? I don't want to copy the block processing code into each function. In other words, is there a way of specifying which filter to call (and with what parameters) as a parameter of the block_process() call?
2. Is there a better way of coding this?

推荐答案

你可以做到像这样：

def block_process(a, blocksize, filt, args):
    b = numpy.empty(a.shape)
    for row in xrange(0, a.shape[0], blocksize):
        for col in xrange(0, a.shape[1], blocksize):
            b[row:row + blocksize, col:col + blocksize] = (
                filt(a[row:row + blocksize, col:col + blocksize], *args))
    return b

无需更正图像右下边缘的不完整块 - 这将自动进行。您可以简单地传入过滤器函数和参数元组。要调用过滤器 filter1（a，filtsize），请使用

There is no need to correct for incomplete blocks at the right and lower edge of the image -- this will happen automatically. You can simply pass in the filter function and the tuple of arguments. To call the filter filter1(a, filtsize), use

block_process(a, blocksize, filter1, (filtsize,))

上面的代码假定第一个参数为过滤器是要过滤的数组，过滤器返回相同形状的过滤数组。

The code above assumes that the first parameter to a filter is the array to be filtered and that a filter returns a filtered array of the same shape.

也可以通过以下方式重写过滤器：它们不会使用尽可能多的内存，因此块处理将变得不必要。

It is also possible that your filters can be rewritten in a way that they don't use as much memory, so the blockprocessing would become unnecessary.

这篇关于从逐块处理函数中调用过滤器函数的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！