저조도 영상 색 복원 기법

영상개선기법(저조도 영상 색복원) under-exposed image enhancement

- 영상의 구조적으로 어두운 부분과 밝은부분을 개선한다. (색감을 유지해야 한다)

 

1. statistical information-based approach

-stretch the dynamic range using histograms

 

2.Retinex-based approach

image can be decomposed into illumination and reflectance

 

2번을 더 많이 사용하고 있다. 

 

 -Single-scale Retinex output(SSR)

 -Multi-scale Retinex output(MSR)

 

---matrix 기반 optimazion을 진행한 기법들

Decomposition and enhancement 

-Bright-pass filter based decomposition

   1. Bright-pass filter baesed decomposition

   2. Variational framework for Retinex

   3. Compute L and R simultaneously

   4. Preserver overall structure and smooth textural details simultaneously

   5. Principal energy-based decomposition

   6. Max pooling on the diffusion space

 

 

Deep neural network-based generation

  -Concept of image translation

  =Residual convolutional neural network with content, color, ...

 

 1. Deep photo enhancer: GAN-based approach

 2. Deep illumination estimation: 중간에 illumination을 한번 예측하자.

 

Technical review of image enhancement

-Retinex-based decomposition has been widely emploted

-Estimated illuminations are adjusted for visibility

-Deep learning-based schemes also have been studied

-Various VQMs can be employed for performance evaluation