Three performance parameters are important to decide the quality analysis on original and reconstructed images (noisy or blurred etc), they are PSNR, MSE and SSIM. Apart from this to decide the saturation and contrast you may also consider standard deviation and mean of an image.
Have you looked into focus stacking to detect the in-focus regions in your image and extract them? It is typically used for multiple images to select the most in focus objects, but you could modify the approach to select features from a single image. I have included to references that do a pretty good job explaining what happens to get you started.