A New Image Structural Similarity Metric Based on K-L Transform

2014 ◽  
pp. 159-168
Author(s):  
Cheng Jiang ◽  
Fen Xiao ◽  
Xiaobo He
Keyword(s):  
Structural Similarity ◽  
Similarity Metric

scholarly journals The Level Weighted Structural Similarity Loss: A Step Away from MSE

2019 ◽  
Vol 33 ◽  
pp. 9989-9990 ◽  
Author(s):  
Yingjing Lu
Keyword(s):  
Mean Square Error ◽  
Structural Similarity ◽  
Generative Models ◽  
Spatial Relationships ◽  
Mean Square ◽  
Model Reconstruction ◽  
Image Domain ◽  
Similarity Metric ◽  
The Mean

The Mean Square Error (MSE) has shown its strength when applied in deep generative models such as Auto-Encoders to model reconstruction loss. However, in image domain especially, the limitation of MSE is obvious: it assumes pixel independence and ignores spatial relationships of samples. This contradicts most architectures of Auto-Encoders which use convolutional layers to extract spatial dependent features. We base on the structural similarity metric (SSIM) and propose a novel level weighted structural similarity (LWSSIM) loss for convolutional Auto-Encoders. Experiments on common datasets on various Auto-Encoder variants show that our loss is able to outperform the MSE loss and the Vanilla SSIM loss. We also provide reasons why our model is able to succeed in cases where the standard SSIM loss fails.

Adaptive Image Steganography Based on Structural Similarity Metric

2011 ◽  
pp. 454-471 ◽  
Author(s):  
Guangjie Liu ◽  
Shiguo Lian ◽  
Yuewei Dai ◽  
Zhiquan Wang
Keyword(s):  
Dynamic Programming ◽  
Information Hiding ◽  
Structural Similarity ◽  
Image Steganography ◽  
The Other ◽  
Trade Off ◽  
Similarity Metric ◽  
Adaptive Quantization ◽  
The Future ◽  
Distortion Index

Image steganography is a common form of information hiding which embeds as many message bits into images and keep the introduced distortion imperceptible. How to balance the trade-off between the capacity and imperceptibility has become a very important issue in the researches of steganography. In this chapter, we discuss one kind of the solution for disposing the trade-off, named adaptive image steganography. After a brief review, we present two methods based on structural similarity metric. The first one is based on the generalized LSB, in which the substitution depth vector is obtained via the dynamic programming under the constraint of an allowable distortion. The second method is proposed to use adaptive quantization-embedder to carry message bits. Different from the first method, the distortion index is constructed by contrast-correlation distortion. The other difference is that the parameters of the adaptive quantization embedder are embedded into the image containing message bits by the reversible da a hiding method. Beside that, we also bring forward some attractive directions worthy of being studied in the future. Furthermore, we find that the existing methods do not have a good way to control the amount of information and the distortion as an extract manner, and most schemes are designed just according to the experiences and experiments.

Sign in / Sign up

Export Citation Format

Share Document