ValueError異常：矩陣不對準

我應該被寫入一個for循環，其執行以下操作：使用奇異向量對應於所述最大的n個奇異值（VRT等的行烏爾等的列和）創建新R，相同尺寸與原始圖像（500×375）ValueError異常：矩陣不對準

這裏的G和B矩陣是我到目前爲止

from PIL import Image 
from Image import new 
from numpy import * 
import numpy as np 
from scipy.linalg import svd 

r, g, b = im.split() 
R = np.array(r.getdata()) 
R = np.asmatrix(R) 
R = np.reshape(R, (375, 500), order = 'F') 
G = np.array(g.getdata()) 
G = np.asmatrix(G) 
G = np.reshape(G, (375, 500), order = 'F') 
B = np.array(b.getdata()) 
B = np.asmatrix(B) 
B = np.reshape(B, (375, 500), order = 'F') 
Ur, Sr, VrT = svd(R.T, full_matrices=False) 
Ug, Sg, VgT = svd(G.T, full_matrices=False) 
Ub, Sb, VbT = svd(R.T, full_matrices=False) 
R1 = np.dot(Ur, diag(Sr)) 
R1 = np.dot(R1, VrT) 
G1 = np.dot(Ug, diag(Sg)) 
G1 = np.dot(G1, VgT) 
B1 = np.dot(Ub, diag(Sb)) 
B1 = np.dot(B1, VbT) 
R1 = np.around([R1]) 
G1 = np.around([G1]) 
B1 = np.around([B1]) 
R1 = np.uint8(R1) 
G1 = np.uint8(G1) 
B1 = np.uint8(B1) 
R1 = R1.T 
G1 = G1.T 
B1 = B1.T 
R1 = R1.flatten('F') 
G1 = G1.flatten('F') 
B1 = B1.flatten('F') 
R1 = tuple(R1) 
G1 = tuple(G1) 
B1 = tuple(B1) 
zipped = zip(R1,G1,B1) 
newim = im.putdata(zipped,1,0) 
im.show(newim) 

for i in xrange(5): 
N = array([200,100,50,10,1]) 
newUr = Ur[0:N[i], : ] 
newSr = newSr[0:N[i]] 
newVrT = VrT[ 0:N[i], :] 
newUg = Ug[0:N[i], : ] 
newSg = Sg[0:N[i]] 
newVgT = VgT[ 0:N[i], :] 
newUb = Ub[0:N[i], : ] 
newSb = Sb[0:N[i]] 
newVbT = VbT[ 0:N[i],:] 
newR = dot(dot(newUr, diag(newSr), newVrT)) 
newG = dot(dot(newUg, diag(newSg), newVgT)) 
newB = dot(dot(newUb, diag(newSb), newVbT)) 
zipped = zip(newR,newG,newB) 
newim = im.putdata(zipped,1,0) 
im.show() 
i = i+1

來源

2012-11-13 Chris C

U，S，和VT的形狀（500375）（375）和（375375） –

我們被要求做S上對角矩陣，其中，N * n項有價值觀和所有其他條目到底newU.shape 0 –

應該是（500，N）newS.shape是（N，N）和newVT.shape是（N，375） –

你可以找到n個最大的值S使用np.argsort。例如，

In [31]: S = np.array([1,3,5,2,4,7]) 

In [32]: np.argsort(S)[-3:] 
Out[32]: array([4, 2, 5]) 

In [33]: idx = np.argsort(S)[-3:] 

In [34]: S[idx] 
Out[34]: array([4, 5, 7])

import Image 
import numpy as np 
linalg = np.linalg 
N = 10 

def ngreatest(arr, n): 
    idx = np.argsort(arr)[-n:] 
    return idx 

img = Image.open(filename).convert('RGB') 
arr = np.asarray(img) 
r, g, b = np.rollaxis(arr, axis = -1) 

Ur, Sr, VrT = linalg.svd(r, full_matrices=False) 
idx = ngreatest(Sr, N) 
Sr = np.diag(Sr[idx]) 
VrT = VrT[idx] 
Ur = Ur[:,idx] 

print(Ur.shape, Sr.shape, VrT.shape)

來源

2012-11-13 02:46:03 unutbu

我開始只是爲了還是應該遵循什麼？ –

我如何找到最小的那麼如果這給了我最大的n值？ –

ValueError異常：矩陣不對準

回答

相關問題