OpenCVのSVMを利用して学習を行う。

# -*- coding: utf-8 -*
import glob
import cv2
import numpy as np

# generateimage array
def creeate_image_array(img_paths):
    images = []
    for img_path in img_paths:
        image = cv2.imread(img_path)
        reshaped_image = image.reshape(image.shape[0] * image.shape[1] * 3)
        images.append(reshaped_image)
    return np.array(images, np.float32)

def main():
    # positive and negative image paths
    pos_img_paths = glob.glob('./data/positive/*')
    neg_img_paths = glob.glob('./data/negative/*')

    # load images
    positive_images = creeate_image_array(pos_img_paths)
    negative_images = creeate_image_array(neg_img_paths)
    images = np.r_[positive_images, negative_images]

    # generate labels
    positive_labels = np.ones(len(pos_img_paths), np.int32)
    negative_labels = np.zeros(len(neg_img_paths), np.int32)
    labels = np.array([np.r_[positive_labels, negative_labels]])

    # SVM
    svm = cv2.ml.SVM_create()
    svm.setType(cv2.ml.SVM_C_SVC)
    svm.setKernel(cv2.ml.SVM_RBF)
    svm.setGamma(10)
    svm.setC(10)
    svm.setTermCriteria((cv2.TERM_CRITERIA_COUNT, 100, 1.e-06))
    svm.train(images, cv2.ml.ROW_SAMPLE, labels)
    svm.save('svm_trained_data.xml')

if __name__ == '__main__':
    main()

無加工の画像を利用して学習を行っている。
出力した学習モデルを利用して分類をテストするコードは下記のようになる。

# -*- coding: utf-8 -*
import glob
import cv2
import numpy as np

# generate image array
def creeate_image_array(img_paths):
    images = []
    for img_path in img_paths:
        image = cv2.imread(img_path)
        reshaped_image = image.reshape(image.shape[0] * image.shape[1] * 3)
        images.append(reshaped_image)
    return np.array(images, np.float32)

def main():
    # set test target images
    test_img_paths = glob.glob('./data/test_tagert/*')

    # loading
    svm = cv2.ml.SVM_load("svm_trained_data.xml")
    
    # test
    images = creeate_image_array(test_img_paths)
    predicted = svm.predict(hist)
    result = predicted[1]
    for (i, img_path) in enumerate(test_img_paths):
        if result[i][0] == 1.0:
            print("yes", img_path)
        else:
            print("no", img_path)

if __name__ == '__main__':
    main()

参考

https://github.com/opencv/opencv/blob/master/samples/python/letter_recog.py#L62
https://algorithm.joho.info/programming/python/hog-svm-classifier-py/