image_bytes

This project is for having low-level control of reading and writing raw and yuv images. It contains a collection of methods that are useful for interacting with the raw bytes from images. If you don't care to deal at the technical level you may be better served by trying rawpixels.net first. opencv now also supports converting a wide variety of bayer and yuv images.

Examples

Convert a IYUV to rgb

from image_bytes.image_bytes import *

yuv = file2arr('data/image_720x1280.yuv', 720, 1280, bpp=1, rowwise=True, yuv_channels=True)
y, u, v = arr2yuv(yuv)
color = yuv2rgb(y, u, v)
cv2.imwrite('output/image_720x1280.jpg', color[:, :, ::-1])

Convert YUV NV12 to rgb

yuv = file2arr('data/image_1664x1664.yuv', 1664, 1664, bpp=1, rowwise=True, yuv_channels=True)
y, u, v = arr2yuv(yuv, interleaved=True)
cv2.imwrite('output/test.jpg', u)
color = yuv2rgb(y, u, v)
cv2.imwrite('output/image_1664x1664.jpg', color[:, :, ::-1])

Y channel only

This yuv image has only the Y-channel:

y = file2arr('data/video_1080x1920.yuv', 1080, 1920, bpp=1, rowwise=True, yuv=False)
cv2.imwrite('output/1080x1920.jpg', y)

Y channel only video

Since it is a video you can read multiple frames:

read_yuv_video('data/video_1080x1920.yuv', 'output/1080x1920_frame_', 1080, 1920, bpp=1, rowwise=True, yuv_channels=False, header_len=0)

Y channel header

Sometimes video files have a header for each frame that must also be skipped. This is an example of skipping a header that contains meta data such as the height, width, frame number, format, etc.

y = file2arr('data/video_1080x1920_h40.yuv', 1080, 1920, offset=40, bpp=1, rowwise=True, yuv_channels=False)
cv2.imwrite('output/1080x1920.jpg', y)

Y channel header video

Reading video frames while skipping headers:

read_yuv_video('data/video_1080x1920_h40.yuv', 'output/1080x1920_h40_frame_', 1080, 1920, bpp=1, rowwise=True, yuv_channels=False, header_len=40)

Convert RAW HDR to rgb

Using a classic freeman demosaic, I've had better luck with other images, but this one is at least recognizable. Probably need to tune some parameters, but putting in as is for now.

bayer = file2arr('data/hdr4_raw_2402x4700_4864_pck_10bits.raw', 2402, 4700, bpp=2, rowwise=False)
color_bayer = flat2channels(bayer, offsets=[[[0,0]],[[1,0],[0,1]],[[1,1]]], blocksize=(2,2))
color = freeman_demosaic(color_bayer, bits=10)
cv2.imwrite('output/hdr_raw.jpg', color[:, :, ::-1])

Note: The bayer pattern is two bytes, if you want to save as an image, I recommend using the PNG format which supports 16 bit images. When reading in the image you will need extra parameters.

cv2.imwrite("output/bayer.png", bayer)
bayer = cv2.imread("output/bayer.png", cv2.IMREAD_GRAYSCALE + cv2.IMREAD_ANYDEPTH)

Using OpenCV to convert

OpenCV now has functions that take care of YUV conversion.

yuv = file2arr('data/image_1664x1664.yuv', 1664, 1664, bpp=1, rowwise=True, yuv_channels=True)
bgr= cv2.cvtColor(yuv, cv2.COLOR_YUV2BGR_NV12)
cv2.imwrite('output/opencv_1664x1664.jpg', bgr)