CODEWITHSUNDEEP
python-3.ximagenumpyopencvstreaming
CodeFlan
CodeFlan

Reputation: 115

Reading RGB raw input, converting to openCV object, then converting to .JPEG bytes - without PIL

I am struggling to read raw RGB image stream from buffer, converting to openCV (numpy) array and encoding back to .jpg Bytes without using PIL. I can achieve this using PIL, but can't figure out how to achieve the same just using numpy/opencv:

Reading in RBG raw stream with PIL (works):

PILFr = Image.frombuffer('RGB',(768,576),buf,"raw",'RGB',0,1)
opencvPILFr = np.array(PILFr)

Attempting a similar read with openCV, I have:

FlatNp=np.frombuffer(buf,dtype=np.uint8, count=-1)
opencvPILFr=np.resize(FlatNp,(576,768))

re-encoding to buffer (self.frame) as JPEG bytes stream with PIL (works):

PILFr=Image.fromarray(cv2.cvtColor(opencvPILFr, cv2.COLOR_BGR2RGB))
tmp = io.BytesIO()
PILFr.save(tmp,format='jpeg', quality=100)
self.frame = tmp.getvalue()

Attempting to re-encode with openCV (does not work!):

tmp = io.BytesIO()
tmp=cv2.imencode(".jpg",opencvPILFr) #this returns a tuple? not numpy array?
self.frame=tmp.tobytes()

Help appreciated -I can see some reasons why the above WOULDN'T work...but can't see how to fix it.

Upvotes: 2

Views: 2352

Answers (1)

CodeFlan
CodeFlan

Reputation: 115

I figured this out. Further improvements welcome. Still not entirely sure I understand the shape and contents of the imencode returned tuple.

See below my working numpy/opencv alternative to the PIL approach above:

Read in RBG 24bit image from buffer and convert to numpy array (opencvFr) for openCV:

FlatNp=np.frombuffer(buf,dtype=np.uint8, count=-1)
opencvFr=np.resize(FlatNp,(576,768,3))

Encoding the edited numpy image array (opencvFr) as JPEG bytes:

tmp=cv2.imencode('.jpg',opencvFr) #this returns a tuple.
self.frame=tmp[1].tobytes() #bytes containing compressed JPEG image

(I am experimenting with this for a streaming camera application)

Upvotes: 1

Related Questions