[python-opencv] PNG 裁切物体

拿到一组图PNG的图，边缘有点太宽了，需要裁切一下，为了这个需求，简单复习一下基本语法。

1.  读取PNG的4个通道

image = cv.imread(image_path, cv.IMREAD_UNCHANGED)

2. 遍历文件夹获取每个图像文件：

def traverse_folder(folder_path):
    for root, dirs, files in os.walk(folder_path):
        for file in files:
            # 文件的绝对路径
            file_path = os.path.join(root, file)
            print(file_path)

3. 找到PNG中物体的ROI：

def findROI(img_path):
    image = cv.imread(img_path)
    gray = cv.cvtColor(image, cv.COLOR_BGR2GRAY)
    # 获取图像的宽高通道
    height, width, channels = image.shape
    # 进行边缘检测
    edges = cv.Canny(gray, 50, 255)

    xmin = width - 1
    ymin = height - 1
    xmax = 0
    ymax = 0
    for i in range(width):
        for j in range(height):
            if edges[j, i] != 0:
                if xmin > i:
                    xmin = i
                if xmax < i:
                    xmax = i
                if ymin > j:
                    ymin = j
                if ymax < j:
                    ymax = j

    print("width : ", xmax - xmin)
    print("height : ", ymax - ymin)
    return xmin,xmax,ymin,ymax

4. 遍历文件夹将jpg转PNG

def traverse_folder_convertJPGtoPNG(folder_path, output_path):
    for root, dirs, files in os.walk(folder_path):
        for file in files:
            # 文件的绝对路径
            file_path = os.path.join(root, file)
            print(file_path)
            # 获取没有后缀的文件名
            filename_without_extension = os.path.splitext(file)[0]
            print(filename_without_extension)
            #print(output_path + '/' + filename_without_extension + '.png')
            jpg_to_png(file_path, output_path + '/' + filename_without_extension + '.png')

def jpg_to_png(jpg_path, png_path):
    img = Image.open(jpg_path)  # 打开jpg文件
    img.save(png_path, 'PNG')  # 保存为png文件

5. 按照指定ROI图像获取图像区域并保存

def cropbysize_and_save_image(image_path, output_path, outputWidth,outputHeight):
    image = cv.imread(image_path, cv.IMREAD_UNCHANGED)
    height, width, channels = image.shape
    print("图像宽度:", width)
    print("图像高度:", height)
    print("图像频道:", channels)
    ymin=math.floor(height*0.5-outputHeight*0.5)
    ymax=math.ceil(height*0.5+outputHeight*0.5)
    xmin=math.floor(width*0.5-outputWidth*0.5)
    xmax=math.ceil(width*0.5+outputWidth*0.5)

    cropped_image = image[ymin:ymax, xmin:xmax]
    # 保存为png文件
    cv.imwrite(output_path, cropped_image, [int(cv.IMWRITE_PNG_STRATEGY_DEFAULT), 9])  

6. 按照问题所在ROI获取图像区域

def crop_and_save_image(image_path, output_path):
    image = cv.imread(image_path, cv.IMREAD_UNCHANGED)
    gray = cv.cvtColor(image, cv.COLOR_BGR2GRAY)
    height, width, channels = image.shape
    edges = cv.Canny(gray, 50, 255)
    print(edges)
    xmin = width - 1
    ymin = height - 1
    xmax = 0
    ymax = 0
    for i in range(width):
        for j in range(height):
            if edges[j,i] != 0:
                if xmin > i:
                   xmin = i
                if xmax < i:
                   xmax = i
                if ymin > j:
                   ymin = j
                if ymax < j:
                   ymax = j

    cropped_image = image[ymin-2:ymax+2, xmin-2:xmax+2]
    cv.imwrite(output_path,cropped_image, [int(cv.IMWRITE_PNG_STRATEGY_DEFAULT), 9]) 

Imwrite参数说明