class BFrameHandle(): def __init__(self, QP): self.data = [] self.IMT = ImageTransform() self.IT = IntegerTransform() self.IT.QuantizationMatrix(QP) def encode3Channels(self, IFrame, PFrame, BFrame): yuvI = self.IMT.rgb2yuv(IFrame) yuvP = self.IMT.rgb2yuv(PFrame) yuvB = self.IMT.rgb2yuv(BFrame) [Y_I, Cr_I, Cb_I] = self.IMT.chromaSub(yuvI) [Y_P, Cr_P, Cb_P] = self.IMT.chromaSub(yuvP) [Y_B, Cr_B, Cb_B] = self.IMT.chromaSub(yuvB) [diffY, motionInfoY] = self.encode(Y_I,Y_P,Y_B) [diffCr, motionInfoCr] = self.encode(Cr_I,Cr_P,Cr_B) [diffCb, motionInfoCb] = self.encode(Cb_I,Cb_P,Cb_B) return [diffY, motionInfoY, diffCr, motionInfoCr, diffCb, motionInfoCb] def encode(self, IFrame, PFrame, BFrame): mvB = MotionVecB(IFrame, PFrame, BFrame) #Initialize A instance #[motionVect1, minMad1, motionVect2, minMad2] motionInfo = mvB.getTwoMotionVector() estimatedBFrame = mvB.recoverPfromI(IFrame, PFrame, motionInfo) diffEstMinusReal = BFrame - estimatedBFrame # Integer Transfer diffEstMinusRealVec = self.IMT.vecMat(diffEstMinusReal, 4) diffEstMinusRealVecIntTran = self.IT.EnIntegerTransformVec(diffEstMinusRealVec) return [diffEstMinusRealVecIntTran, motionInfo] def decode3Channels(self, IFrame, PFrame, diffAndmotionVector): yuvI = self.IMT.rgb2yuv(IFrame) yuvP = self.IMT.rgb2yuv(PFrame) [Y_I, Cr_I, Cb_I] = self.IMT.chromaSub(yuvI) [Y_P, Cr_P, Cb_P] = self.IMT.chromaSub(yuvP) [diffY, motionInfoY, diffCr, motionInfoCr, diffCb, motionInfoCb] = diffAndmotionVector Y_B = self.decode(Y_I, Y_P, diffY, motionInfoY) Cr_B = self.decode(Cr_I, Cr_P, diffCr, motionInfoCr) Cb_B = self.decode(Cb_I, Cb_P, diffCb, motionInfoCb) #Expand all 3 channels yuvRec = self.IMT.chromaExpand(Y_B, Cr_B, Cb_B) rgbImRec = self.IMT.yuv2rgb(yuvRec) rgbIm = self.IMT.double2uintImage(rgbImRec) return rgbIm def decode(self, IFrame, PFrame, diff, motionInfo): mvB = MotionVecB(IFrame, PFrame, np.zeros_like(IFrame)) #Here use both I frame to initialize, as no need for Pframe estimatedBFrame = mvB.recoverPfromI(IFrame, PFrame, motionInfo) #DeInteger Transform for diff diffDetraned = self.IT.DeIntegerTransformVec(diff) diffRec = self.IMT.dvecMat(np.shape(IFrame), diffDetraned, 4) PFrame = estimatedBFrame + diffRec return PFrame
import cv2
from MotionVec import *
from IntegerTransform import *
from ImageTransform import *
from FrameHandleHelpers import *
from IFrameHandle import *
#frames = []
QP = 0
cycleCount = 0
CONSTANT_VIDEO_PATH = "SampleVideo_360x240_50mb.mp4"
cap = cv2.VideoCapture(CONSTANT_VIDEO_PATH)
IMT = ImageTransform()
IT = IntegerTransform()
IT.QuantizationMatrix(0)
FRameBuffer = []
Frame = []
FrameDecode = []
PHand = PFrameHandle(QP)
IHand = IFrameHandle(QP)
'''
Displaying Sequence: I P P I P P I P P
Coding Sequence I P B B P B B I B B
'''
while True:
if cycleCount % 3 == 0: #IFrame
ret, Frame = cap.read()
class IntraCoding(): """This class is for intra coding and decoding for I frames""" def __init__(self,QP): #self.deintra_image = [] self.intra_image = [] self.IT = IntegerTransform() self.IT.QuantizationMatrix(QP) def IntraEncoding(self,Y,Y_r,i,j,N): rows = i - 1 cols = j - 1 T = Y_r[rows,cols+1:cols+N+1] L = Y_r[rows+1:rows+N+1,cols] c0 = self.mode0(Y,T,N) c1 = self.mode1(Y,L,N) c2 = self.mode2(Y,L,T,N) BK = Y[rows+1:rows+1+N,cols+1:cols+1+N] SOD = [0,0,0] SOD[0] = sum(sum(abs(c0-BK))) SOD[1] = sum(sum(abs(c1-BK))) SOD[2] = sum(sum(abs(c2-BK))) mode = SOD.index(min(SOD)) if (mode == 0): self.intra_image[i:i+4,j:j+4] = c0 return BK - c0 elif (mode == 1): self.intra_image[i:i+4,j:j+4] = c1 return BK - c1 elif (mode == 2): self.intra_image[i:i+4,j:j+4] = c2 return BK - c2 '''def IntraDecoding(self,Y,mode,i,j,N): rows = i - 1 cols = j - 1 for k in range(0,2*N): T(k) = Y(rows,cols+k) L(k) = Y(rows+k,cols) if (mode == 0): self.deintra_image(i:i+4,j:j+4) = mode0(Y,T,N) elif (mode == 1): self.deintra_image(i:i+4,j:j+4) = mode1(Y,L,N) elif (mode == 2): self.deintra_image(i:i+4,j:j+4) = mode2(Y,L,T,N) elif (mode == 3): self.deintra_image(i:i+4,j:j+4) = np.zeros_like(Y(0:4,0:4))''' def IntraCodingVec(self,Y): self.intra_image = np.zeros_like(Y) Y_r = np.zeros_like(Y) for m in range(np.shape(Y)[0]//16): for n in range(np.shape(Y)[1]//16): for i in range(m*4,m*4+4): for j in range(n*4,n*4+4): BK = Y[i*4:i*4+4,j*4:j*4+4] if (i == 0 and j == 0): mode = 3 ICP_en = self.IT.EnIntegerTransform(BK) Y_r[i*4:i*4+4,j*4:j*4+4] = self.IT.DeIntegerTransform(ICP_en) elif (j == 0): rows = i - 1 T = Y_r[rows,0:4] mode = 0 self.intra_image[i*4:i*4+4,j*4:j*4+4] = self.mode0(Y,T,4) ICP = BK - self.intra_image[i*4:i*4+4,j*4:j*4+4] ICP_en = self.IT.EnIntegerTransform(ICP) ICP_r = self.IT.DeIntegerTransform(ICP_en) Y_r[i*4:i*4+4,j*4:j*4+4] = self.intra_image[i*4:i*4+4,j*4:j*4+4] + ICP_r elif (i == 0): cols = j - 1 L = Y_r[0:4,cols] mode = 1 self.intra_image[i*4:i*4+4,j*4:j*4+4] = self.mode1(Y,L,4) ICP = BK - self.intra_image[i*4:i*4+4,j*4:j*4+4] ICP_en = self.IT.EnIntegerTransform(ICP) ICP_r = self.IT.DeIntegerTransform(ICP_en) Y_r[i*4:i*4+4,j*4:j*4+4] = self.intra_image[i*4:i*4+4,j*4:j*4+4] + ICP_r else: ICP = self.IntraEncoding(Y,Y_r,i*4,j*4,4) ICP_en = self.IT.EnIntegerTransform(ICP) ICP_r = self.IT.DeIntegerTransform(ICP_en) Y_r[i*4:i*4+4,j*4:j*4+4] = self.intra_image[i*4:i*4+4,j*4:j*4+4] + ICP_r return Y_r '''def IntraDeCodingVec(self,Y): self.deintra_image = np.zeros_like(Y) for m in range(np.shape(Y)[0]//16): for n in range(np.shape(Y)[1]//16): for i in range(m*4,m*4+4): for j in range(n*4,n*4+4): self.IntraDeCoding(Y,self.mode(i,j),i*4,j*4,4)''' def mode0(self,Y,T,N): out = np.zeros_like(Y[0:4,0:4]) for i in range(N): for j in range(N): out[i,j] = T[i] return out def mode1(self,Y,L,N): out = np.zeros_like(Y[0:4,0:4]) for i in range(N): for j in range(N): out[i,j] = L[i] return out def mode2(self,Y,L,T,N): out = np.zeros_like(Y[0:4,0:4]) for i in range(N): for j in range(N): out[i,j] = round(np.mean(np.concatenate([L,T]))+4/9.0) return out