基于OpenCV的图像分割是怎样的

import cv2 import matplotlib.pyplot as plt import numpy as np import scipy.misc import scipy.ndimage import skimage.filters import sklearn.metrics # Turn on interactive mode. Turn off with plt.ioff() plt.ion()

grayscale = scipy.misc.imread('grayscale.png')grayscale = 255 - grayscalegroundtruth = scipy.misc.imread('groundtruth.png')plt.subplot(1, 3, 1)plt.imshow(255 - grayscale, cmap='gray')plt.title('grayscale')plt.axis('off')plt.subplot(1, 3, 2)plt.imshow(grayscale, cmap='gray')plt.title('inverted grayscale')plt.axis('off')plt.subplot(1, 3, 3)plt.imshow(groundtruth, cmap='gray')plt.title('groundtruth binary')plt.axis('off')

median_filtered = scipy.ndimage.median_filter(grayscale, size=3) plt.imshow(median_filtered, cmap='gray') plt.axis('off') plt.title("median filtered image")

counts, vals = np.histogram(grayscale, bins=range(2 ** 8)) plt.plot(range(0, (2 ** 8) - 1), counts) plt.title("Grayscale image histogram") plt.xlabel("Pixel intensity") plt.ylabel("Count")

result = skimage.filters.thresholding.try_all_threshold(median_filtered)

threshold = skimage.filters.threshold_otsu(median_filtered) print("Threshold value is {}".format(threshold)) predicted = np.uint8(median_filtered > threshold) * 255 plt.imshow(predicted, cmap='gray') plt.axis('off') plt.title("otsu predicted binary image")

def get_confusion_matrix_elements(groundtruth_list, predicted_list):    """returns confusion matrix elements i.e TN, FP, FN, TP as floats  See example code for helper function definitions    """    _assert_valid_lists(groundtruth_list, predicted_list)
    if _all_class_1_predicted_as_class_1(groundtruth_list, predicted_list) is True:        tn, fp, fn, tp = 0, 0, 0, np.float64(len(groundtruth_list))
    elif _all_class_0_predicted_as_class_0(groundtruth_list, predicted_list) is True:        tn, fp, fn, tp = np.float64(len(groundtruth_list)), 0, 0, 0
    else:        tn, fp, fn, tp = sklearn.metrics.confusion_matrix(groundtruth_list, predicted_list).ravel()        tn, fp, fn, tp = np.float64(tn), np.float64(fp), np.float64(fn), np.float64(tp)
    return tn, fp, fn, tp

def get_accuracy(groundtruth_list, predicted_list):
    tn, fp, fn, tp = get_confusion_matrix_elements(groundtruth_list, predicted_list)        total = tp + fp + fn + tn    accuracy = (tp + tn) / total        return accuracy

def get_f1_score(groundtruth_list, predicted_list):    """Return f1 score covering edge cases"""
    tn, fp, fn, tp = get_confusion_matrix_elements(groundtruth_list, predicted_list)        if _all_class_0_predicted_as_class_0(groundtruth_list, predicted_list) is True:        f1_score = 1    elif _all_class_1_predicted_as_class_1(groundtruth_list, predicted_list) is True:        f1_score = 1    else:        f1_score = (2 * tp) / ((2 * tp) + fp + fn)
    return f1_score

def get_mcc(groundtruth_list, predicted_list):    """Return mcc covering edge cases"""   
    tn, fp, fn, tp = get_confusion_matrix_elements(groundtruth_list, predicted_list)        if _all_class_0_predicted_as_class_0(groundtruth_list, predicted_list) is True:        mcc = 1    elif _all_class_1_predicted_as_class_1(groundtruth_list, predicted_list) is True:        mcc = 1    elif _all_class_1_predicted_as_class_0(groundtruth_list, predicted_list) is True:        mcc = -1    elif _all_class_0_predicted_as_class_1(groundtruth_list, predicted_list) is True :        mcc = -1
    elif _mcc_denominator_zero(tn, fp, fn, tp) is True:        mcc = -1
    # Finally calculate MCC    else:        mcc = ((tp * tn) - (fp * fn)) / (            np.sqrt((tp + fp) * (tp + fn) * (tn + fp) * (tn + fn)))        return mcc

>>> validation_metrics = get_validation_metrics(groundtruth, predicted) {'mcc': 0.8533910225863214, 'f1_score': 0.8493358633776091, 'tp': 5595.0, 'fn': 1863.0, 'fp': 122.0, 'accuracy': 0.9924278259277344, 'tn': 254564.0}

def get_confusion_matrix_intersection_mats(groundtruth, predicted):    """ Returns dict of 4 boolean numpy arrays with True at TP, FP, FN, TN    """
    confusion_matrix_arrs = {}
    groundtruth_inverse = np.logical_not(groundtruth)    predicted_inverse = np.logical_not(predicted)
    confusion_matrix_arrs['tp'] = np.logical_and(groundtruth, predicted)    confusion_matrix_arrs['tn'] = np.logical_and(groundtruth_inverse, predicted_inverse)    confusion_matrix_arrs['fp'] = np.logical_and(groundtruth_inverse, predicted)    confusion_matrix_arrs['fn'] = np.logical_and(groundtruth, predicted_inverse)
    return confusion_matrix_arrs

def get_confusion_matrix_overlaid_mask(image, groundtruth, predicted, alpha, colors):    """    Returns overlay the 'image' with a color mask where TP, FP, FN, TN are    each a color given by the 'colors' dictionary    """    image = cv2.cvtColor(image, cv2.COLOR_GRAY2RGB)    masks = get_confusion_matrix_intersection_mats(groundtruth, predicted)    color_mask = np.zeros_like(image)    for label, mask in masks.items():        color = colors[label]        mask_rgb = np.zeros_like(image)        mask_rgb[mask != 0] = color        color_mask += mask_rgb    return cv2.addWeighted(image, alpha, color_mask, 1 - alpha, 0)
alpha = 0.5confusion_matrix_colors = {   'tp': (0, 255, 255),  #cyan   'fp': (255, 0, 255),  #magenta   'fn': (255, 255, 0),  #yellow   'tn': (0, 0, 0)     #black   }validation_mask = get_confusion_matrix_overlaid_mask(255 - grayscale, groundtruth, predicted, alpha, confusion_matrix_colors)print('Cyan - TP')print('Magenta - FP')print('Yellow - FN')print('Black - TN')plt.imshow(validation_mask)plt.axis('off')plt.title('confusion matrix overlay mask')