Morphological Operations

Preface

This tutorial applies to firmware version: CanMV_K230_YAHBOOM_micropython_v1.8-13. Download from the official Canaan website at https://download.kendryte.com/developer/releases/canmv_k230_micropython/daily_build/

Example Results

Run the example code for this section: 02.camera_morphology.py.

This section first converts the camera color image to grayscale, then generates a binary image through OTSU automatic thresholding, and finally demonstrates erosion, dilation, opening, closing, morphological gradient, top hat, and black hat operations.

Results are shown below

Principles

What Are Morphological Operations

Morphological operations use a small matrix sliding over the image, called a structuring element or convolution kernel. It mainly processes white foreground in binary images. The shape and size of the kernel determine the neighborhood range.

• Erosion: White areas shrink, can remove small white spots.
• Dilation: White areas expand, can connect nearby areas.
• Opening: Erosion followed by dilation, commonly used to remove small white noise.
• Closing: Dilation followed by erosion, commonly used to fill small black holes and connect breaks.
• Morphological Gradient: Dilation result minus erosion result, highlights contours.
• Top Hat: Original image minus opening result, highlights small bright areas.
• Black Hat: Closing result minus original image, highlights small dark areas.

Structuring Elements

OpenCV supports rectangular, cross, and elliptical structuring elements. This example pre-allocates 5×5 rectangular and elliptical kernels:

KS = 5
kr = cv2.getStructuringElement(cv2.MORPH_RECT, (KS, KS))
ke = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (KS, KS))

Creating kernels outside the loop avoids repeated array allocation per frame.

Code Overview

Import Modules

xxxxxxxxxx
import time, os, sys, gc
from media.sensor import *
from media.display import *
from media.media import *
import cv2
import ulab.numpy as np

Initialize Camera and Display Module

xxxxxxxxxx
W, H = 640, 480
sensor = Sensor(width=1280, height=960, fps=90)
sensor.reset()
sensor.set_framesize(width=W, height=H)
sensor.set_pixformat(Sensor.RGB888)
Display.init(Display.ST7701, width=640, height=480, to_ide=True)
sensor.run()
time.sleep(0.5)

Set Morphological Mode

xxxxxxxxxx
mode = 2

Grayscale and Binarization Preprocessing

xxxxxxxxxx
img = sensor.snapshot()
img_np = img.to_numpy_ref()
gray = cv2.cvtColor(img_np, cv2.COLOR_BGR2GRAY)
ret, binary = cv2.threshold(gray, 0, 255, cv2.THRESH_OTSU)

OTSU automatically selects the threshold based on the current frame's grayscale histogram. ret saves the calculated threshold, and binary is a binary image containing only 0 and 255.

Execute Morphological Operations

xxxxxxxxxx
if mode == 0:
    out = cv2.erode(binary, kr)
elif mode == 1:
    out = cv2.dilate(binary, kr)
elif mode == 2:
    out = cv2.morphologyEx(binary, cv2.MORPH_OPEN, ke)
elif mode == 3:
    out = cv2.morphologyEx(binary, cv2.MORPH_CLOSE, ke)
elif mode == 4:
    out = cv2.morphologyEx(binary, cv2.MORPH_GRADIENT, ke)
elif mode == 5:
    out = cv2.morphologyEx(binary, cv2.MORPH_TOPHAT, ke)
else:
    out = cv2.morphologyEx(binary, cv2.MORPH_BLACKHAT, ke)

Erosion and dilation use rectangular kernels, while other operations use elliptical kernels. This way you can observe basic operations and compare the effects of different shaped structuring elements on edges.

Wrap Grayscale Image and Display

xxxxxxxxxx
img_out = image.Image(
    W, H, image.GRAYSCALE,
    alloc=image.ALLOC_REF,
    data=out
)
Display.show_image(img_out)

ALLOC_REF directly references the out array without copying the full image. The array must remain valid until display is complete.

Resource Release

xxxxxxxxxx
finally:
    sensor.stop()
    Display.deinit()
    os.exitpoint(os.EXITPOINT_ENABLE_SLEEP)
    time.sleep_ms(100)

Parameter Adjustment Notes

• Kernel size KS: Must be a positive number, typically odd numbers like 3, 5, 7. Larger kernel means stronger effect but higher computation.
• Foreground polarity: Erosion and dilation are defined for white foreground. If the target is black in the binary image, the visual effect will be opposite to expectations. You can use THRESH_BINARY_INV instead.
• OTSU stability: Illumination changes cause the automatic threshold to vary per frame. For stable contrast, you can fix the illumination or use a fixed threshold.
• Number of iterations: The current call executes only once. For stronger effect, you can pass iterations, but start from 1 and increase gradually.
• Performance: Continuous morphological operations at 640×480 generate temporary arrays. If frame rate is insufficient, you can first drop to 320×240 to compare effects.