如何让Python模拟的蛋白质扩散图像更具真实感？

当前基于随机游走+高斯平滑+随机噪声的方案，生成的图像Blob形态过于规整、亮斑缺乏连续性，可通过以下针对性调整让模拟更贴近真实生物成像效果：

1. 用受限聚集游走替代纯随机游走 #

真实蛋白质扩散会受细胞内结构阻碍，且倾向于聚集，让粒子移动时带有黏附概率和聚集倾向，生成的密度分布更不规则：

import numpy as np
from scipy.ndimage import gaussian_filter

def aggregating_random_walk(n_particles, steps, canvas_size=32, stick_prob=0.1, agg_prob=0.3):
    canvas = np.zeros((canvas_size, canvas_size))
    positions = np.random.randint(0, canvas_size, (n_particles, 2))
    
    for _ in range(steps):
        # 计算邻域密度，引导聚集
        density = gaussian_filter(canvas, sigma=1)
        moves = []
        for (x, y) in positions:
            if np.random.rand() < agg_prob:
                # 优先向密度更高的邻域移动
                neighbors = [(nx, ny) for nx in [x-1,x,x+1] for ny in [y-1,y,y+1] 
                             if 0<=nx<canvas_size and 0<=ny<canvas_size]
                best_move = neighbors[np.argmax([density[nx, ny] for nx, ny in neighbors])]
                moves.append(np.array(best_move) - (x, y))
            else:
                moves.append(np.random.choice([-1,0,1], size=2))
        moves = np.array(moves)
        new_pos = positions + moves
        new_pos = np.clip(new_pos, 0, canvas_size-1)
        # 黏附概率：粒子有概率停留原地，强化局部密集区
        stick_mask = np.random.rand(n_particles) < stick_prob
        new_pos[stick_mask] = positions[stick_mask]
        # 更新画布
        for (x, y) in new_pos:
            canvas[x, y] += 1
        positions = new_pos
    return canvas

2. 多尺度混合平滑替代单一高斯滤波 #

单一高斯卷积会让Blob边缘过于圆润，混合不同尺度的高斯滤波，并叠加局部最大值增强，模拟真实聚集区的不均匀亮斑：

from scipy.ndimage import maximum_filter

def multi_scale_enhancement(canvas):
    # 混合不同标准差的高斯滤波，保留不同尺度的聚集特征
    smooth_small = gaussian_filter(canvas, sigma=1.2)
    smooth_large = gaussian_filter(canvas, sigma=2.5)
    mixed_smooth = 0.6 * smooth_small + 0.4 * smooth_large
    # 局部最大值增强，突出连续亮斑
    local_peaks = maximum_filter(mixed_smooth, size=3)
    enhanced = np.where(mixed_smooth == local_peaks, mixed_smooth * 1.5, mixed_smooth)
    return enhanced

3. 结构化噪声替代纯随机噪声 #

生物成像的背景噪声并非完全随机，用低分辨率纹理噪声叠加泊松噪声，更贴近真实成像的异质性背景：

def structured_background_noise(canvas, noise_scale=0.05):
    # 生成低分辨率噪声再放大，模拟背景纹理
    low_res_noise = np.random.randn(8, 8)
    background_texture = np.kron(low_res_noise, np.ones((4, 4)))
    # 泊松噪声符合生物成像的光子统计特性
    poisson_noise = np.random.poisson(canvas * 0.1)
    # 混合噪声并归一化强度范围
    noisy_canvas = canvas + background_texture * noise_scale + poisson_noise
    return np.clip(noisy_canvas / noisy_canvas.max(), 0, 1)

完整流程示例 #

# 生成聚集游走密度图
canvas = aggregating_random_walk(n_particles=50, steps=200)
# 多尺度增强
enhanced = multi_scale_enhancement(canvas)
# 添加结构化噪声
final_image = structured_background_noise(enhanced)

# 可视化
import matplotlib.pyplot as plt
plt.imshow(final_image, cmap='hot')
plt.axis('off')
plt.show()

内容的提问来源于stack exchange，提问作者hexaquark