Python implements license plate positioning and segmentation

Specific steps

1. Convert the collected color license plate image into a grayscale image
2. Use Gaussian smoothing on the grayscale image, and then perform medium straight filtering on it
3. Use the Sobel operator to perform edge detection on the image.
4. Perform erosion, expansion, opening and closing morphological combination transformation on the binary image.
5. Perform morphological transformation on the image. Perform contour search and extract the license plate according to its aspect ratio

Code implementation

image grayscale

gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

Gaussian smoothing, median filtering

gaussian = cv2.GaussianBlur(gray, (3, 3), 0, 0, cv2.BORDER_DEFAULT)
median = cv2.medianBlur(gaussian, 5)

##Sobel edge detection

sobel = cv2.Sobel(median, cv2.CV_8U, 1, 0,  ksize = 3)

Binarization

ret, binary = cv2.threshold(sobel, 170, 255, cv2.THRESH_BINARY)

The form of erosion, expansion, opening and closing operations on the binary image Learn combination transformation

##Perform contour search on the morphologically transformed image, and extract the license plate according to its aspect ratio

1. Search License plate area

# 膨胀和腐蚀操作的核函数
element1 = cv2.getStructuringElement(cv2.MORPH_RECT, (9, 1))
element2 = cv2.getStructuringElement(cv2.MORPH_RECT, (8, 6))
# 膨胀一次，让轮廓突出
dilation = cv2.dilate(binary, element2, iterations = 1)
# 腐蚀一次，去掉细节
erosion = cv2.erode(dilation, element1, iterations = 1)
# 再次膨胀，让轮廓明显一些
dilation2 = cv2.dilate(erosion, element2,iterations = 3)

2. Use green lines to draw the license plate area and cut the license plate

def findPlateNumberRegion(img):
    region = []
    # 查找轮廓
    contours,hierarchy = cv2.findContours(img, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)

    # 筛选面积小的
    for i in range(len(contours)):
        cnt = contours[i]
        # 计算该轮廓的面积
        area = cv2.contourArea(cnt)

        # 面积小的都筛选掉
        if (area < 2000):
            continue

        # 轮廓近似，作用很小
        epsilon = 0.001 * cv2.arcLength(cnt,True)
        approx = cv2.approxPolyDP(cnt, epsilon, True)

        # 找到最小的矩形，该矩形可能有方向
        rect = cv2.minAreaRect(cnt)
        print "rect is: "
        print rect

        # box是四个点的坐标
        box = cv2.cv.BoxPoints(rect)
        box = np.int0(box)

        # 计算高和宽
        height = abs(box[0][1] - box[2][1])
        width = abs(box[0][0] - box[2][0])

        # 车牌正常情况下长高比在2.7-5之间
        ratio =float(width) / float(height)
        if (ratio > 5 or ratio < 2):
            continue

        region.append(box)

    return region

The above is the detailed content of Python implements license plate positioning and segmentation. For more information, please follow other related articles on the PHP Chinese website!