안녕하세요. 허세임입니다. 오늘은 친구의 10년전 사진을 가지고 사진어플없이 chatgpt와의 대화로 사진을 복원해보는 작업을 해보겠습니다.

시작하기

먼저 오늘의 준비물 사진입니다.

대한민국 남자아이들이라면 모두 하나씩 가지고있는 태권도 사진입니다. 색이 많이 바래있네요. 이래서 사진은 디지털로 소장해야하나봅니다.

이번작업은 gpt4 ADA로 진행하였습니다.

먼저 이 사진의 디테일에 대해 설명하여, gpt가 이미지를 이해하는데 도움을 줍니다.

from PIL import Image
import IPython.display as display

# Load the image
image_path = '/mnt/data/IMG_7115.jpeg'
original_image = Image.open(image_path)

# Display the image
display.display(original_image)

import cv2
import numpy as np

# Convert the image to a numpy array
np_image = np.array(original_image)

# Convert the image from RGB to BGR because OpenCV uses BGR format
np_image = cv2.cvtColor(np_image, cv2.COLOR_RGB2BGR)

# First, we'll apply histogram equalization to improve the contrast of the image
# This is usually a good first step in the process of image restoration
# We need to convert the image to a LAB color space, equalize the L channel and then convert back to RGB
lab = cv2.cvtColor(np_image, cv2.COLOR_BGR2LAB)
l, a, b = cv2.split(lab)
l_eq = cv2.equalizeHist(l)
lab_eq = cv2.merge((l_eq, a, b))
np_image_eq = cv2.cvtColor(lab_eq, cv2.COLOR_LAB2BGR)

# Now we'll convert it back to RGB for further processing
np_image_eq = cv2.cvtColor(np_image_eq, cv2.COLOR_BGR2RGB)

# Convert back to PIL Image and display the result
restored_image = Image.fromarray(np_image_eq)
display.display(restored_image)

# Save the intermediate result to the disk
intermediate_result_path = '/mnt/data/IMG_7115_restored_intermediate.jpeg'
restored_image.save(intermediate_result_path)

intermediate_result_path

전후 비교 사진입니다. 이미지가 좀 선명해진것같긴한데, 전체적으로 푸르스름한 보라빛 제거가 필요해보입니다.

# Function to reduce blue and purple tones
def reduce_blue_purple_tones(image, strength=1.0):
    """
    Reduce the blue and purple tones in the image.
    
    :param image: PIL Image to be processed
    :param strength: Factor by which the blue and purple tones are reduced
    :return: PIL Image with reduced blue and purple tones
    """
    # Convert the image to RGB if it's not already in that mode
    if image.mode != 'RGB':
        image = image.convert('RGB')
    
    # Split into separate channels
    r, g, b = image.split()
    
    # Reduce blue channel
    b = b.point(lambda i: i * (1 - strength))
    
    # Merge channels back and convert to RGB
    result_image = Image.merge('RGB', (r, g, b))
    
    return result_image

# Apply the function to reduce blue and purple tones
strength_of_reduction = 0.5  # Adjust the strength as needed
final_restored_image = reduce_blue_purple_tones(restored_image, strength=strength_of_reduction)

# Display the final result
display.display(final_restored_image)

# Save the final result to the disk
final_result_path = '/mnt/data/IMG_7115_restored_final.jpeg'
final_restored_image.save(final_result_path)

final_result_path