Convert 1D 16bit image data to 2D numpy array

Question

I have a 1D array of size 18874568 created from bytes as shown below

np_img = np.array(list(data), dtype=np.uint16)

How can I convert this to a 2D array of size 3072 x 3072 and in 8bit?

The bytes data is acquired from a image capture device called as a flat panel detector (FPD). It's specified in the docs that the FPD captures 16bit image of size 3072x3072.

16bit raw image is attached below https://drive.google.com/file/d/1Kw1UeKOaBGtXNpxGsCXEk-gjGw3YaDJJ/view?usp=sharing

Edit: C# code on conversion given by the FPD support team

    private Bitmap Load16bppGrayImage(string fileName, int offset=0)
    {
        Bitmap loadind_bmp;
        var raw = File.ReadAllBytes(fileName).Skip(offset).ToArray();
        var pwi = 3072;
        var phi = 3072;
        var dpiX = 96d;
        var dpiY = 96d;

        loadind_bmp = new Bitmap(pwi, phi, PixelFormat.Format48bppRgb);

        BitmapData bmpData = loadind_bmp.LockBits(new Rectangle(0, 0, loadind_bmp.Width, loadind_bmp.Height), ImageLockMode.ReadWrite, loadind_bmp.PixelFormat);
        int depth = Bitmap.GetPixelFormatSize(bmpData.PixelFormat) / 16;
        IntPtr srcAdr = bmpData.Scan0;

        unsafe
        {
            for (int y = 0; y < bmpData.Height; y++)
            {
                int pos = y * bmpData.Stride / 3;
                ushort* imgPtr = (y * bmpData.Stride / 2) + (ushort*)bmpData.Scan0;

                for (int x = 0; x < bmpData.Width; x++)
                {
                    ushort value = (ushort)(raw[pos + (x * sizeof(ushort)) + 1] << 8 | raw[pos + x * sizeof(ushort)]);

                    imgPtr[x * depth] = value;    //Blue 0-255
                    imgPtr[x * depth + 1] = value;   //Green   0-255
                    imgPtr[x * depth + 2] = value;  //Red 0-255
                }
            }
        }

        loadind_bmp.UnlockBits(bmpData);
        return loadind_bmp;

    }

The data you've posted is 18874568 bytes long; sqrt(18874568/2) is > 3072, so there has to be some other data than just raw image data in there (18874568-(3072*3072*2) = 200 bytes to be exact). — AKX
– AKX, Commented Jun 8, 2022 at 7:51
Msybe you can show a screengrab of what parameters you input into IrfanView? — Mark Setchell
– Mark Setchell, Commented Jun 8, 2022 at 8:12
@AKX So, now we are short of data... we need 3072*3072*3*2 bytes. — Mark Setchell
– Mark Setchell, Commented Jun 8, 2022 at 18:38
The Fpd image is combined by the header(96 bytes) + image data (Width * Height * 2) + extra infomation(100 bytes) + CRC(4 bytes) - Their reply — Arki99
– Arki99, Commented Jun 9, 2022 at 7:03

AKX · Accepted Answer · 2022-06-09 08:06:30Z

1

Right, based on all of the information gathered by the comments, I think you'll want something like this:

import numpy as np
import PIL.Image as Image

with open("data1.raw", "rb") as fp:
    data = fp.read()

w = h = 3072
dt = np.dtype(np.uint16).newbyteorder("<")  # little-endian
image = np.frombuffer(data, offset=96, count=w * h, dtype=dt).reshape((h, w))

# Dump to 16bpp TIFF:
Image.fromarray(image).save("data1.tiff")

The resulting image is quite dark, not "just white" – maybe it needs to be inverted? Who knows... :)

edited Jun 9, 2022 at 8:06

answered Jun 9, 2022 at 7:45

AKX

171k16 gold badges146 silver badges228 bronze badges

Sign up to request clarification or add additional context in comments.

3 Comments

Mark Setchell Over a year ago

Actually, I was just thinking np.fromfile() with offset and count might be a fraction cleaner than open() and read()... Don't get me wrong, it's good already ;-)

AKX Over a year ago

@MarkSetchell Sure, but OP says they have a data already. :)

Mark Setchell Over a year ago

D'oh! I better go put the kettle on - I clearly need more coffee!

Mechanic Pig · Accepted Answer · 2022-06-08 07:48:15Z

0

Use ndarray.view:

>>> a = np.random.randint(2 ** 7, 2 ** 12, 10, dtype=np.uint16)
>>> a
array([3434, 1565, 3144,  357, 3755, 3130, 3799,  338, 3224, 2794],
      dtype=uint16)
>>> a.view(np.uint8)
array([106,  13,  29,   6,  72,  12, 101,   1, 171,  14,  58,  12, 215,
        14,  82,   1, 152,  12, 234,  10], dtype=uint8)

The better way is to read with uint8 instead of uint16 before conversion. You should refer to: Input and output

edited Jun 8, 2022 at 7:48

answered Jun 8, 2022 at 7:30

Mechanic Pig

7,7813 gold badges17 silver badges35 bronze badges

1 Comment

AKX Over a year ago

Couldn't you just reinterpret an uint16 matrix as an uint8 array without any manual shifting?

Collectives™ on Stack Overflow

Convert 1D 16bit image data to 2D numpy array

2 Answers 2

3 Comments

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

3 Comments

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related