satomic/screendetection

nothing

ScreenDetection

In some cases, we need to use a phone to photo another phone's screen, in order to detect the quality of the screen. So there several problems need to be solved.

---

brief functions introduce

• basic functions

  • get_sharp_index(image, contour=None) how sharp of an image, region only is supported
  • get_normalized_3_color_distribution(image) return tuple with 3 normalized number, refer to Blue/Green/Red
  • get_color_spectrum(image)

• cut out contour region from an image

  • get_max_rectangle_contour(image, unpefect_ratio=0.1, area_ratio=0.02, debug=False)
  • get_region_by_contour(image, contour)
  • get_resized_region_by_contour(image, contour, height=800, width=480)

• whether the camera is out of focus

  • is_clear(image, sharp_index=0.01, contour=None)

• how the colur is shifted from image A to B in channel BGR

  • BGR_shift_A_2_B(image_a, image_b)

• compare two photos, return a similarity index

  • compare_images_ssim(image_a, image_b) this is fucking from from skimage.measure import compare_ssim, but this need the two images have the same shape, it's so fucking bad. so we need a more awsome function, judge them by the color spectrum

---

realtime demo

run the example_realtime.py you can get this

---

example

with original image is below

• 3 color distribution: [ 0.56661608 0.58174488 0.58354015]
• sharp: 0.043625642481, this value is more than 0.01, that means the image is clear, the camera is not out of focus when shooting

---

by catch the max rectangle (similar) in the image, we have the image blow

---

if we fill the contour, then we get a mask, by apply the mask to the original image, we have the region.jpg

• region.jpg sharp with black backgroud: 0.00546950920615, we can see the value is very smaller then the orignal image sharp, cuz the details except the screen are all lost, and the sharp caculation algrathm is edges_pixels/image_pixels, so to get a more availabe sharp index, we need to minimize the image_pixels, only contain the contour area. luckily, the function is already exists.
• region.jpg sharp within contour: 0.0388608524458

---

in real case, we wanna the resized screen of the phone, this can be called "rebuild", by using the function I mentioned upper. we get the resized screen below

• 3 color distribution: [ 0.76739759 0.63802456 0.06344758], we can easily find that compare to the original image, the B channel shift more, and the R channel is almost disappear
• resized_region sharp: 0.0228776041667

---

so how the color shift from the orignal image to resized region image, we can use the funtion BGR_shift_A_2_B,positive channel means color increased, the bigger, the more increased

• color shift: [ 0.20078152 0.05627968 -0.52009258], just like we have ananlysised upper