pip install numpy opencv-python pillow

import numpy as np
import cv2
from PIL import Image
import matplotlib.pyplot as plt

def load_and_preprocess(image_path):
    """加載圖片并進(jìn)行預(yù)處理"""
    # 使用OpenCV讀取圖片
    image = cv2.imread(image_path)
    # 轉(zhuǎn)換為灰度圖
    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
    # 歸一化到0-1范圍
    normalized = gray.astype(np.float32) / 255.0
    return normalized

# 示例使用
image = load_and_preprocess('your_image.jpg')

def calculate_gradient_features(image):
    """計(jì)算圖像的梯度特征"""
    # 使用Sobel算子計(jì)算x和y方向的梯度
    sobelx = cv2.Sobel(image, cv2.CV_64F, 1, 0, ksize=3)
    sobely = cv2.Sobel(image, cv2.CV_64F, 0, 1, ksize=3)
    
    # 計(jì)算梯度幅度和方向
    magnitude = np.sqrt(sobelx**2 + sobely**2)
    direction = np.arctan2(sobely, sobelx)
    
    return magnitude, direction

def detect_rotation_angle(image):
    """檢測(cè)圖片的旋轉(zhuǎn)角度"""
    # 計(jì)算四個(gè)可能角度的得分
    scores = []
    possible_angles = [0, 90, 180, 270]
    
    for angle in possible_angles:
        # 旋轉(zhuǎn)圖片
        if angle == 0:
            rotated = image
        else:
            # 使用NumPy的rot90函數(shù)進(jìn)行旋轉(zhuǎn)
            k = angle // 90
            rotated = np.rot90(image, k)
        
        # 計(jì)算水平方向的梯度差異（正常方向的圖片應(yīng)該有明顯的水平梯度）
        horizontal_grad = np.abs(cv2.Sobel(rotated, cv2.CV_64F, 1, 0, ksize=3))
        vertical_grad = np.abs(cv2.Sobel(rotated, cv2.CV_64F, 0, 1, ksize=3))
        
        # 計(jì)算得分：水平梯度應(yīng)該大于垂直梯度
        score = np.sum(horizontal_grad) / (np.sum(vertical_grad) + 1e-6)
        scores.append(score)
    
    # 選擇得分最高的角度
    best_angle = possible_angles[np.argmax(scores)]
    return best_angle

# 使用示例
angle = detect_rotation_angle(image)
print(f"檢測(cè)到的旋轉(zhuǎn)角度: {angle}度")

def optimized_rotation_detection(image):
    """優(yōu)化版的旋轉(zhuǎn)檢測(cè)"""
    # 預(yù)計(jì)算所有旋轉(zhuǎn)版本
    rotations = [
        image,  # 0度
        np.rot90(image, 1),  # 90度
        np.rot90(image, 2),  # 180度
        np.rot90(image, 3)   # 270度
    ]
    
    # 使用向量化計(jì)算所有旋轉(zhuǎn)版本的梯度
    gradients = [np.abs(cv2.Sobel(rot, cv2.CV_64F, 1, 0, ksize=3)) for rot in rotations]
    
    # 計(jì)算所有得分
    scores = [np.sum(grad) for grad in gradients]
    
    return [0, 90, 180, 270][np.argmax(scores)]

def matrix_based_detection(image):
    """基于矩陣運(yùn)算的檢測(cè)方法"""
    # 計(jì)算圖像的矩（moments）
    moments = cv2.moments(image)
    
    # 計(jì)算重心
    if moments['m00'] != 0:
        cx = moments['m10'] / moments['m00']
        cy = moments['m01'] / moments['m00']
    else:
        cx, cy = 0, 0
    
    # 計(jì)算中心矩
    mu20 = moments['mu20'] / moments['m00']
    mu02 = moments['mu02'] / moments['m00']
    mu11 = moments['mu11'] / moments['m00']
    
    # 計(jì)算方向角度
    angle = 0.5 * np.arctan2(2 * mu11, mu20 - mu02)
    angle_deg = np.degrees(angle)
    
    # 將角度映射到最接近的90度倍數(shù)
    possible_angles = [0, 90, 180, 270]
    closest_angle = min(possible_angles, key=lambda x: abs(x - angle_deg))
    
    return closest_angle

def correct_image_rotation(image_path):
    """自動(dòng)校正圖片旋轉(zhuǎn)"""
    # 加載圖片
    image = cv2.imread(image_path)
    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
    
    # 檢測(cè)旋轉(zhuǎn)角度
    angle = detect_rotation_angle(gray)
    
    # 進(jìn)行旋轉(zhuǎn)校正
    if angle == 90:
        corrected = cv2.rotate(image, cv2.ROTATE_90_CLOCKWISE)
    elif angle == 180:
        corrected = cv2.rotate(image, cv2.ROTATE_180)
    elif angle == 270:
        corrected = cv2.rotate(image, cv2.ROTATE_90_COUNTERCLOCKWISE)
    else:
        corrected = image
    
    return corrected, angle

# 使用示例
corrected_image, detected_angle = correct_image_rotation('rotated_image.jpg')
cv2.imwrite('corrected_image.jpg', corrected_image)

def batch_process_images(image_paths):
    """批量處理多張圖片"""
    results = []
    
    for path in image_paths:
        try:
            corrected, angle = correct_image_rotation(path)
            output_path = f"corrected_{path.split('/')[-1]}"
            cv2.imwrite(output_path, corrected)
            results.append((path, angle, output_path))
        except Exception as e:
            print(f"處理圖片 {path} 時(shí)出錯(cuò): {e}")
    
    return results

def process_uploaded_image(uploaded_file):
    """處理用戶(hù)上傳的圖片"""
    # 將上傳的文件轉(zhuǎn)換為NumPy數(shù)組
    image_array = np.frombuffer(uploaded_file.read(), np.uint8)
    image = cv2.imdecode(image_array, cv2.IMREAD_COLOR)
    
    # 轉(zhuǎn)換為灰度圖用于檢測(cè)
    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
    
    # 檢測(cè)旋轉(zhuǎn)角度
    angle = optimized_rotation_detection(gray)
    
    # 如果需要旋轉(zhuǎn)，進(jìn)行校正
    if angle != 0:
        if angle == 90:
            image = cv2.rotate(image, cv2.ROTATE_90_CLOCKWISE)
        elif angle == 180:
            image = cv2.rotate(image, cv2.ROTATE_180)
        elif angle == 270:
            image = cv2.rotate(image, cv2.ROTATE_90_COUNTERCLOCKWISE)
    
    return image, angle

class ImageRotationProcessor:
    """圖片旋轉(zhuǎn)處理器類(lèi)"""
    
    def __init__(self):
        self.rotation_counts = {0: 0, 90: 0, 180: 0, 270: 0}
    
    def process_directory(self, directory_path):
        """處理整個(gè)目錄中的圖片"""
        import os
        results = []
        
        for filename in os.listdir(directory_path):
            if filename.lower().endswith(('.png', '.jpg', '.jpeg')):
                path = os.path.join(directory_path, filename)
                try:
                    corrected, angle = correct_image_rotation(path)
                    output_path = os.path.join(directory_path, f"corrected_{filename}")
                    cv2.imwrite(output_path, corrected)
                    
                    self.rotation_counts[angle] += 1
                    results.append((filename, angle, output_path))
                    
                except Exception as e:
                    print(f"處理 {filename} 時(shí)出錯(cuò): {e}")
        
        return results
    
    def get_stats(self):
        """獲取處理統(tǒng)計(jì)信息"""
        return self.rotation_counts

# 使用示例
processor = ImageRotationProcessor()
results = processor.process_directory('./images/')
stats = processor.get_stats()
print(f"處理統(tǒng)計(jì): {stats}")

def enhance_contrast(image):
    """增強(qiáng)圖片對(duì)比度"""
    # 使用直方圖均衡化
    enhanced = cv2.equalizeHist(image)
    return enhanced

def detect_with_contrast_enhancement(image_path):
    """帶對(duì)比度增強(qiáng)的旋轉(zhuǎn)檢測(cè)"""
    image = cv2.imread(image_path, cv2.IMREAD_GRAYSCALE)
    enhanced = enhance_contrast(image)
    angle = detect_rotation_angle(enhanced)
    return angle

def detect_rotation_with_confidence(image):
    """帶置信度檢測(cè)的旋轉(zhuǎn)判斷"""
    scores = []
    possible_angles = [0, 90, 180, 270]
    
    for angle in possible_angles:
        if angle == 0:
            rotated = image
        else:
            k = angle // 90
            rotated = np.rot90(image, k)
        
        horizontal_grad = np.abs(cv2.Sobel(rotated, cv2.CV_64F, 1, 0, ksize=3))
        vertical_grad = np.abs(cv2.Sobel(rotated, cv2.CV_64F, 0, 1, ksize=3))
        
        score = np.sum(horizontal_grad) / (np.sum(vertical_grad) + 1e-6)
        scores.append(score)
    
    best_idx = np.argmax(scores)
    best_angle = possible_angles[best_idx]
    best_score = scores[best_idx]
    
    # 計(jì)算置信度
    sorted_scores = sorted(scores, reverse=True)
    confidence = (sorted_scores[0] - sorted_scores[1]) / sorted_scores[0] if sorted_scores[0] > 0 else 0
    
    return best_angle, confidence

# 使用示例
angle, confidence = detect_rotation_with_confidence(image)
if confidence > 0.3:  # 置信度閾值
    print(f"檢測(cè)到旋轉(zhuǎn)角度: {angle}度, 置信度: {confidence:.2f}")
else:
    print("無(wú)法確定旋轉(zhuǎn)角度，圖片可能沒(méi)有明確方向")