model1-2.py

#coding=gbk
import zipfile
import cStringIO as sio
import scipy.misc as misc
import numpy as np

def s2o(s):
    return sio.StringIO(s)

dataset = zipfile.ZipFile('glasses.zip','r')
noglasslist = [i for i in dataset.namelist() if i.startswith('alignedNoneGlasses') and i.endswith('bmp')]
glasslist = [i for i in dataset.namelist() if i.startswith('alignedGlasses') and i.endswith('bmp')]

def liner():
    "提取眼睛行（带眼镜）"
    #import os
    #try: os.mkdir('glassline')
    #except: pass
    import PIL.Image
    glassmodel = np.empty((len(glasslist),(70-25)*(90-0)),np.uint8)
    idx = 0
    for i in glasslist:
        img = PIL.Image.open(s2o(dataset.read(i)))
        img=img.crop((0,25,90,70))
        glassmodel[idx] = misc.fromimage(img).flatten()
        #img.save('glassline\\'+i.split('/')[-1])
        print i
        idx+=1
    print glassmodel.shape
    np.save('glassline.npy',glassmodel)

    nglassmodel = np.empty((len(noglasslist),(70-25)*(90-0)),np.uint8)
    idx = 0
    for i in noglasslist:
        img = PIL.Image.open(s2o(dataset.read(i)))
        img=img.crop((0,25,90,70))
        nglassmodel[idx] = misc.fromimage(img).flatten()
        #img.save('glassline\\'+i.split('/')[-1])
        print i
        idx+=1
    print nglassmodel.shape
    np.save('nglassline.npy',nglassmodel)

def sccodedirect():
    "得到不带眼镜的RPCA结果"
    nglassmodel = np.load('nglassline.npy').astype('f')
    from sklearn.decomposition import SparsePCA
    learning = SparsePCA(500,verbose=True)
    learning.fit(nglassmodel)
    import cPickle
    cPickle.dump(learning,file('sparsepcadirect','wb'),-1)

def scdecomp():
    "对带眼镜照片进行分解，得到还原表示以及眼镜区域"
    glassmodel = np.load('glassline.npy').astype('f')
    import cPickle
    sparsedirect = cPickle.load(file('sparsepcadirect','rb'))
    #sparsedirect.set_params(transform_algorithm='lasso_lars')
    dout = sparsedirect.transform(glassmodel)
    drec = dout.dot(sparsedirect.components_)
    np.save('glassdecomppca.npy',drec)
    np.save('glassdiffpca.npy',glassmodel - drec)

def scvisdecomp():
    "可视化结果"
    from layerbase import DrawPatch
    import cPickle
    sparsedirect = cPickle.load(file('sparsepcadirect','rb'))
    drec = sparsedirect.components_.reshape((-1,1,70-25,90-0))
    misc.toimage(DrawPatch(drec)).save('dict.jpg')
    drec = np.load('glassline.npy').astype('f').reshape((-1,1,70-25,90-0))
    misc.toimage(DrawPatch(drec)).save('src.jpg')
    drec = np.load('glassdecomppca.npy').reshape((-1,1,70-25,90-0))
    misc.toimage(DrawPatch(drec)).save('decomppca.jpg')
    drec = np.abs(np.load('glassdiffpca.npy')).reshape((-1,1,70-25,90-0))
    misc.toimage(DrawPatch(drec)).save('diffpca.jpg')

def cutratio(ratio,imgid,sparsedirect=None, glassmodel=None):
    "使用卡阈值标记损坏区域的方法迭代SparseCoding去除眼镜"
    if sparsedirect==None:
        import cPickle
        sparsedirect = cPickle.load(file('sparsepcadirect','rb'))
        sparsedirect.set_params(transform_algorithm='lasso_lars')
    modelval = sparsedirect.components_
    if glassmodel==None:
        glassmodel = np.load('glassline.npy').astype('f')
    sample = glassmodel[imgid:imgid+1]

    #初始还原损坏区域
    rec = sparsedirect.transform(sample).dot(modelval)
    diff = np.abs(sample - rec)[0]
    diffmax = np.max(diff)

    maskarea = np.where(diff>diffmax*ratio,0,1)
    
    itx = 1
    while True:
        newmodel = np.copy(modelval)
        sparsedirect.components_ = newmodel * maskarea[np.newaxis,:]
        newsample = np.copy(sample)
        newsample = newsample * maskarea[np.newaxis,:]

        rec = sparsedirect.transform(newsample).dot(modelval)
        diff = np.abs(sample - rec)[0]
        diffmax = np.max(diff)
        
        newmaskarea = np.where(diff>diffmax*ratio,0,1)
        maskdiff = np.sum(np.abs(maskarea - newmaskarea))
        print "ITERATION",itx,"DIFF",maskdiff
        itx += 1
        maskarea = newmaskarea
        if maskdiff == 0:
            print "Converge"
            break
    sparsedirect.components_ = modelval
    return rec, diff, maskarea

def cutratio_all(ratio):
    import cPickle
    sparsedirect = cPickle.load(file('sparsepcadirect','rb'))
    sparsedirect.set_params(transform_algorithm='lasso_lars')
    glassmodel = np.load('glassline.npy').astype('f')
    glassmodel = glassmodel[:200]

    recall = np.empty_like(glassmodel)
    diffall = np.empty_like(glassmodel)
    maskall = np.empty_like(glassmodel)

    for idx in range(glassmodel.shape[0]):
        print idx,glassmodel.shape[0]
        recall[idx],diffall[idx],maskall[idx] = cutratio(ratio, idx, sparsedirect, glassmodel)

    np.save('cut_rec.npy',recall)
    np.save('cut_diff.npy',diffall)
    np.save('cut_mask.npy',maskall)
    from layerbase import DrawPatch
    drec = recall.reshape((-1,1,70-25,90-0))
    misc.toimage(DrawPatch(drec)).save('irec.jpg')
    drec = diffall.reshape((-1,1,70-25,90-0))
    misc.toimage(DrawPatch(drec)).save('idiff.jpg')
    drec = maskall.reshape((-1,1,70-25,90-0))
    misc.toimage(DrawPatch(drec)).save('imask.jpg')

if __name__=="__main__":
    print "Make model 1"
    #liner()
    sccodedirect()
    scdecomp()
    scvisdecomp()
    #cutratio_all(0.3)