人脸补全

人脸补全

需求：

根据人的上半边脸预测下半边脸，用各种算法取得的结果与原图比较

思考：

• 这是一个回归问题，不是分类问题（人脸数据不固定）
• 数据集一共包含40个人，每一个人10张照片，分布规律
• 每一个人取出8张照片作为训练数据，2张照片作为测试数据
• 样本特征和样本标签如何拆分？上半边脸作为样本特征，下半边脸作为特征标签

• 效果图

• 导包

import numpy as np
import pandas as pd
from pandas import Series,DataFrame
import matplotlib.pyplot as plt
%matplotlib inline

from sklearn.datasets import fetch_olivetti_faces
from sklearn.neighbors import KNeighborsRegressor
from sklearn.linear_model import LinearRegression,Ridge,Lasso

• 处理数据

faces = fetch_olivetti_faces()
data = faces.data
images = faces.images
target = faces.target

X_train = []
y_train = []
X_test = []
y_test = []

for i in range(40):
    for j in range(10):
        face_data = data[i*10+j]
        up_face = face_data[:2048]
        bottom_face = face_data[2048:]
        if j < 8:
            X_train.append(up_face)
            y_train.append(bottom_face)
        else:
            X_test.append(up_face)
            y_test.append(bottom_face)

• 训练数据

knn = KNeighborsRegressor()
linear = LinearRegression()
ridge = Ridge()
lasso = Lasso()

knn.fit(X_train,y_train)
linear.fit(X_train,y_train)
ridge.fit(X_train,y_train)
lasso.fit(X_train,y_train)

• 预测数据

y_ = knn.predict(X_test)
line_y_ = linear.predict(X_test)
ridge_y_ = ridge.predict(X_test)
lasso_y_ = lasso.predict(X_test)

# 把所有的预测结果保存的一个列表
pre_results = [y_,line_y_,ridge_y_,lasso_y_]
titles = ['True','KNN','Linear','Ridge','Lasso']

• 绘图

# 画布
plt.figure(figsize=(10,10))
for i in range(5):
    # 先取出真实的脸
    true_up_face = X_test[i].reshape(32,64)
    true_bottom_face = y_test[i].reshape(32,64)
    true_face = np.concatenate((true_up_face,true_bottom_face),axis=0)
    axes = plt.subplot(5,5,5*i+1)
    axes.set_title(titles[0])
    axes.axis('off')
    plt.imshow(true_face,cmap='gray')
    for index,y_ in enumerate(pre_results):
        # 获取到每一种算法模型预测出的下半边脸的数据
        pre_bottom_face = y_[i].reshape(32,64)
        pre_face = np.concatenate((true_up_face,pre_bottom_face),axis=0)
        axes = plt.subplot(5,5,5*i+index+2)
        axes.set_title(titles[index+1])
        axes.axis('off')
        plt.imshow(pre_face,cmap='gray')