Python 代码：将语谱图、MFCC 特征图与原始特征图相乘生成新图片

import os
import cv2
import numpy as np

path_spectrograms = r'D:\论文代码\spectrograms'
path_spectrogram = r'D:\论文代码\语谱图'
path_mfcc = r'D:\论文代码\MFCC'
path_output = r'D:\论文代码\feature_multiplication'

# 获取所有子目录
subdirs = []
for root, dirs, files in os.walk(path_spectrograms):
    for d in dirs:
        subdirs.append(os.path.join(root, d))

# 处理每个子目录
for subdir in subdirs:
    # 创建对应的输出目录
    subdir_rel = os.path.relpath(subdir, path_spectrograms)
    output_subdir = os.path.join(path_output, subdir_rel)
    os.makedirs(output_subdir, exist_ok=True)

    # 遍历子目录下的文件
    for filename in os.listdir(subdir):
        # 读取语谱图
        spectrogram_path = os.path.join(subdir, filename)
        spectrogram = cv2.imread(spectrogram_path, cv2.IMREAD_GRAYSCALE)

        # 读取对应的特征图
        spectrogram_rel = os.path.relpath(spectrogram_path, path_spectrograms)
        feature_path = os.path.join(path_spectrogram, spectrogram_rel)
        feature = cv2.imread(feature_path, cv2.IMREAD_GRAYSCALE)
        feature = cv2.resize(feature, spectrogram.shape[::-1])

        # 特征图相乘
        feature = cv2.normalize(feature.astype(np.float32), None, 0, 1, cv2.NORM_MINMAX)
        spectrogram = cv2.normalize(spectrogram.astype(np.float32), None, 0, 1, cv2.NORM_MINMAX)
        result = (feature * spectrogram).astype(np.uint8)

        # 读取对应的MFCC特征图
        mfcc_path = os.path.join(path_mfcc, spectrogram_rel)
        mfcc = cv2.imread(mfcc_path, cv2.IMREAD_GRAYSCALE)
        mfcc = cv2.resize(mfcc, spectrogram.shape[::-1])

        # 特征图相乘
        mfcc = cv2.normalize(mfcc.astype(np.float32), None, 0, 1, cv2.NORM_MINMAX)
        result = (result * mfcc).astype(np.uint8)

        # 保存结果
        output_path = os.path.join(output_subdir, filename)
        cv2.imwrite(output_path, result)