Image Blur Filtering

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 10.camera_filter_blur.py.

This section demonstrates Gaussian, mean, median, bilateral, box filtering, and custom sharpening kernels on a 640×480 live image. Changing the mode allows you to compare the visual effects and frame rates of different filtering methods.

Principle Explanation

Why Filtering is Needed

Image filtering computes new pixel values from neighboring pixels, which can be used for denoising, smoothing, edge preservation, or sharpening. Different methods have varying effects on noise types, edges, and computation load.

Larger filter kernels provide wider smoothing range but also increase computation and temporary array overhead.

Code Overview

Initialize Camera and Display Module

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)

Create Sharpening Kernel and Set Mode

​x
sharpen = np.array([
    [-1.0, -1.0, -1.0],
    [-1.0,  9.0, -1.0],
    [-1.0, -1.0, -1.0]
])
​
mode = 0

The sharpening kernel has a center weight of 9 and the surrounding 8 positions are -1, totaling 1, so it enhances local differences while roughly preserving overall brightness.

Execute Filtering

xxxxxxxxxx
if mode == 0:
    out = cv2.GaussianBlur(img_np, (7, 7), 0)
elif mode == 1:
    out = cv2.blur(img_np, (7, 7))
elif mode == 2:
    out = cv2.medianBlur(img_np, 7)
elif mode == 3:
    out = cv2.bilateralFilter(img_np, 7, 50, 50)
elif mode == 4:
    out = cv2.boxFilter(img_np, -1, (7, 7), normalize=True)
else:
    out = cv2.filter2D(img_np, -1, sharpen)

ddepth=-1 means the output depth matches the input. Out-of-range results in 8-bit images are handled according to API rules. Pay attention to highlight and dark area saturation when sharpening.

Write Back to Original Frame and Display

xxxxxxxxxx
img_np[:, :, :] = out
Display.show_image(img)

The output array is written back to the shared frame for display. If the original image is still needed afterwards, complete related calculations or explicitly copy the required areas before overwriting.

Garbage Collection and Resource Release

xxxxxxxxxx
if fc % 10 == 0:
    gc.collect()
​
finally:
    sensor.stop()
    Display.deinit()
    os.exitpoint(os.EXITPOINT_ENABLE_SLEEP)
    time.sleep_ms(100)

Parameter Adjustment Notes

• Kernel Size: Start with 3×3 or 5×5 for Gaussian, mean, and box filters. 7×7 is the current example value.
• Median Aperture: Should be an odd number greater than 1; larger aperture significantly increases computation.
• Bilateral Filter: Larger sigmaColor means pixels with greater color differences also affect each other; sigmaSpace controls the spatial neighborhood range.
• Sharpening Strength: The current kernel is relatively strong and will also amplify noise. You can reduce the center weight while keeping the kernel weight sum close to 1.
• Frame Rate Comparison: Bilateral and median filtering are usually slower than mean and Gaussian filtering. Test under the same scene and display settings.
• Edge Detection Preprocessing: When you need to reduce noise while preserving edges as much as possible, start with small-kernel Gaussian filtering rather than directly using large-kernel blurring.