Table 11 Ablation study analysis of the HDLIP-SHAR methodology.

From: Automated image inpainting for historical artifact restoration using hybridisation of transfer learning with deep generative models

Methods

PSNR (dB)

Minimum

Maximum

Average

Transformer-GAN + MF + Hybrid Squeezenet CNN (Without Contrast Enhancement and Segmentation process)

59.31

65.73

62.52

Transformer-GAN + CE + Hybrid Squeezenet CNN (Without Median filter and Segmentation process)

59.93

66.50

63.22

Transformer-GAN + MF + CE + Hybrid Squeezenet CNN (Segmentation process)

60.69

67.11

63.90

HDLIP-SHAR (Transformer-GAN with image pre-processing, feature extraction, and segmentation process)

61.42

67.76

64.59

Back to article page