Thresholding

Preface

This tutorial applies to firmware version: CanMV_K230_YAHBOOM_micropython_v1.8-13. You can download it from the Jianan website https://download.kendryte.com/developer/releases/canmv_k230_micropython/daily_build/

Example Results

Run the example code for this section 12.camera_threshold.py.

This section converts the live image to grayscale and demonstrates fixed global threshold, OTSU automatic threshold, adaptive Gaussian threshold, and adaptive mean inverted threshold.

Principle Explanation

Global Threshold

Global threshold uses the same cutoff value for the entire image. Taking binary threshold as an example:

Pixel value > threshold → 255
Pixel value ≤ threshold → 0

Fixed threshold is simple and fast, but when illumination is uneven, bright and dark regions of the same target may be separated.

OTSU Automatic Threshold

OTSU selects the threshold that maximizes inter-class variance based on the grayscale histogram, suitable for scenes where foreground and background form two distinct grayscale peaks. It is still a global threshold and cannot solve severe local shadows.

Adaptive Threshold

Adaptive threshold computes a local neighborhood threshold for each pixel, more suitable for uneven illumination. The Gaussian method weights the neighborhood, while the mean method uses equal-weight averaging.

Code Overview

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 Threshold Mode

xxxxxxxxxx
mode = 1

Convert to Grayscale

xxxxxxxxxx
gray = cv2.cvtColor(img_np, cv2.COLOR_BGR2GRAY)

The threshold function input is a single-channel 8-bit image.

Fixed Threshold and OTSU

xxxxxxxxxx
if mode == 0:
    ret, out = cv2.threshold(
        gray, 127, 255,
        cv2.THRESH_BINARY
    )
elif mode == 1:
    ret, out = cv2.threshold(
        gray, 0, 255,
        cv2.THRESH_OTSU
    )

ret saves the actual threshold. In fixed mode it is usually 127; in OTSU mode it is the value computed from the current frame.

Adaptive Threshold

xxxxxxxxxx
elif mode == 2:
    out = cv2.adaptiveThreshold(
        gray, 255,
        cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
        cv2.THRESH_BINARY,
        11, 2
    )
else:
    out = cv2.adaptiveThreshold(
        gray, 255,
        cv2.ADAPTIVE_THRESH_MEAN_C,
        cv2.THRESH_BINARY_INV,
        15, 3
    )

Block size must be an odd number greater than 1. C is subtracted from the local statistic value to fine-tune the foreground range. Mode 3 uses inverted output, so dark targets become white foreground.

Display and Resource Release

​x
img_out = image.Image(
    W, H, image.GRAYSCALE,
    alloc=image.ALLOC_REF,
    data=out
)
Display.show_image(img_out)
​
finally:
    sensor.stop()
    Display.deinit()
    os.exitpoint(os.EXITPOINT_ENABLE_SLEEP)
    time.sleep_ms(100)

Parameter Adjustment Notes

• Fixed Threshold 127: First observe the pixel range of target and background in the grayscale image, then select a middle value.
• OTSU Usage Conditions: When the scene grayscale distribution does not have an obvious bimodal pattern, the automatic threshold may not be reliable.
• Block Size: Larger is closer to slowly changing local background; smaller adapts better to fine brightness changes but is more prone to noise.
• Constant C: Increasing or decreasing changes the white foreground proportion. It should be adjusted separately after fixing illumination and block size.
• Foreground Polarity: Contours and morphology typically work with white foreground. Choose THRESH_BINARY or THRESH_BINARY_INV based on whether the target is bright or dark.
• Preprocessing: When there is a lot of binary noise, you can apply small-kernel smoothing before thresholding, or perform opening and closing operations after thresholding.