图像处理的边界检测算法

Question

I need an algorithm that can recognize the overall shape of the input image. I tried edge detection and corner detection but they are identifying the dips and curves as well.

If I input an image, the algorithm should detect the points that are higher than the initial point unless it is the last point in the top of the image and for the bottom, the algorithm should identify those points which are lower than the initial one. Through this process the algorithm should be ale to create an overall boundary shape of the image and then return it.

Any suggestions or links?

Input image:

Expected output:

Answer 1

Since you haven't provided any library, I have shown how it can be done using OpenCV.

As mentioned in the comments, finding the convex hull is what you are looking for. A convex shape is one where no interior angles are greater than 180 degrees. For more simple explanation have a look at his blog page

The following code is written in Python using OpenCV:

# read the input image 
img = cv2.imread('red_plot.png')
img2 = img.copy()

# convert to grayscale
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# obtain binary image, such that the red pixels are white
th = cv2.threshold(gray,127,255,cv2.THRESH_BINARY_INV+cv2.THRESH_OTSU)[1]

# find contours (shapes with white pixels)
contours, hierarchy = cv2.findContours(th, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)

contours contains the boundary points of all the shapes found in the binary image. Notice the word boundary , since we only consider the shape of the white region. For this very purpose we use cv2.RETR_EXTERNAL flag while finding contours.

And since the lines in the plot are not connected, we obtain more than one contour. The following snippet collects the points of all the contours in a single variable:

cc = np.array([[0, 0]], dtype=int)
for i, c in enumerate(contours):
    c_modified = c.reshape(len(contours[i]), 2)
    cc = np.concatenate((cc, c_modified), axis = 0)

cc2 = cc[1:]
combined_contour = np.reshape(cc2, (cc2.shape[0], 1, cc2.shape[1]))

combined_contour contains the boundary points of all the red pixels.

# finding convex hull
hull = cv2.convexHull(combined_contour, False)

# draw the result
color = (0, 255, 0)
img2 = cv2.drawContours(img2, [hull], -1, color, 3)
cv2.imshow('Convex hull', img2)
cv2.waitKey(0)