Question

我正在解决一个问题，我需要为图像中的白色区域找到边界框。由于我正在处理真实世界的图片，因此我在RGB值上设置了阈值，我从中创建了一个蒙版，然后对其进行了标记并从中获取了边框的坐标。这是代码。

import numpy as np
from skimage import io, measure

def bin_labelled_img_to_bboxes(bin_image_labelled):
        bboxes = []
        for j in np.unique(bin_image_labelled):
            if j == 0:
                continue
            curr = (bin_image_labelled == j)
            if np.sum(curr) < 50*50:
                continue
            indices = np.nonzero(curr)
            miny = np.min(indices[0])
            minx = np.min(indices[1])
            maxy = np.max(indices[0])
            maxx = np.max(indices[1])
            bboxes.append(((miny, minx), (maxy, maxx)))
        return bboxes


class WhiteSeperator(object):
    def __init__ (self, img_path):
        self.img_path = img_path
        self.img = io.imread(self.img_path)
        self.bin_image_labelled = np.zeros((self.img.shape[0], self.img.shape[1]))
        self.bboxes = []

    def get_bin_labelled_img(self):
        img = self.img
        chan1 = (img[:,:,0] > 200) * (img[:,:,0] <= 255)
        chan2 = (img[:,:,0] > 180) * (img[:,:,0] <= 255)
        chan3 = (img[:,:,0] > 140) * (img[:,:,0] <= 255)

        bin_img = (chan1*chan2*chan3)
        bin_image_labelled = measure.label(bin_img)

        return bin_image_labelled

    def get_white_bboxes(self):
        final_white_bboxes = []
        self.bin_image_labelled = self.get_bin_labelled_img()
        white_bboxes = bin_labelled_img_to_bboxes(self.bin_image_labelled)

        for bbox in white_bboxes:
            width = bbox[1][1]-bbox[0][1]
            height = bbox[1][0]-bbox[0][0]
            if height > 80 and width > 200:
                self.bboxes.append(bbox)
                final_white_bboxes.append(bbox)
        return final_white_bboxes

对于高分辨率图像（3000像素x 2000像素），每个图像大约需要3-11秒。我的假设是每幅图像的时间差异取决于找到的白色边框的数量（此处归咎于bin_labelled_img_to_bboxes函数）由于我必须在视频帧上执行此操作，因此即使3秒也非常慢。可以通过更有效的方式完成上述操作吗？

加快python图像颜色阈值处理/过滤

0 个答案: