Structural Similarity Measure for Color Images

Hassan, Mohammed; Bhagvati, Chakravarthy

doi:10.5120/6169-8590

Cited by 44 publications

(21 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Further, its pixelwise gradient has a simple analytical form and is inexpensive to compute. In this work, we focus on the original grayscale SSIM and MS-SSIM, although there are interesting variations and improvements such as colorized SSIM [19,12].…”

Section: Structural Similaritymentioning

confidence: 99%

Learning to generate images with perceptual similarity metrics

Snell

Ridgeway

Liao

et al. 2017

2017 IEEE International Conference on Image Processing (ICIP)

102

View full text Add to dashboard Cite

Deep networks are increasingly being applied to problems involving image synthesis, e.g., generating images from textual descriptions and reconstructing an input image from a compact representation. Supervised training of image-synthesis networks typically uses a pixel-wise loss (PL) to indicate the mismatch between a generated image and its corresponding target image. We propose instead to use a loss function that is better calibrated to human perceptual judgments of image quality: the multiscale structural-similarity score (MS-SSIM) [34]. Because MS-SSIM is differentiable, it is easily incorporated into gradient-descent learning. We compare the consequences of using MS-SSIM versus PL loss on training deterministic and stochastic autoencoders. For three different architectures, we collected human judgments of the quality of image reconstructions. Observers reliably prefer images synthesized by MS-SSIM-optimized models over those synthesized by PL-optimized models, for two distinct PL measures (L 1 and L 2 distances). We also explore the effect of training objective on image encoding and analyze conditions under which perceptually-optimized representations yield better performance on image classification. Finally, we demonstrate the superiority of perceptually-optimized networks for super-resolution imaging. Just as computer vision has advanced through the use of convolutional architectures that mimic the structure of the mammalian visual system, we argue that significant additional advances can be made in modeling images through the use of training objectives that are well aligned to characteristics of human perception.

show abstract

Section: Structural Similaritymentioning

confidence: 99%

Learning to generate images with perceptual similarity metrics

Snell

Ridgeway

Liao

et al. 2017

2017 IEEE International Conference on Image Processing (ICIP)

102

View full text Add to dashboard Cite

show abstract

“…For the filtering assessment we have used the objective quality measures MAE, PSNR, and DIS k (defined from CMSSIM [39]) to objectively compare the performance of a selected group of filters. These measures are defined as follows [37]:…”

Section: α-Stable Distributionmentioning

confidence: 99%

Fuzzy averaging filter for impulse noise reduction in colour images with a correction step

Gregori¹,

Morillas²,

Sala³

et al. 2018

Journal of Visual Communication and Image Representation

View full text Add to dashboard Cite

In this paper we propose a fuzzy detection and reduction method for impulse noise in colour images. Detection is based on the fuzzyfication of a well-known statistic called ROD. The noise degrees obtained are used to reduce impulses by employing a fuzzy averaging between the input colour vector and a robust estimate of noise-free colour vector within the input neighbourhood. Fuzzy averaging has some advantages in terms of both noise reduction and detail preservation in front of detect and replace approaches because of threshold based decisions of the latter. However, robustness of the former is lower. We solve this problem by including a correction mechanism that checks the fuzzy noise degree of the output and replaces it with a robust colour vector either when noise has not been properly reduced or when a colour artefact has been introduced. We carry out a thorough study of the method parameter setting and give a convenient and robust setting. Experimental results show that our approach is very robust in front of four different types of impulse noise.

show abstract

“…Next, we measure the similarity between all distorted images and the original one with the usual similarity measures MAE, MSE, NCD, as well as with Structural Similarity Index (SSIM) [13,14] (used by averaging after component-wise application in each channel), FSIMc [26], CMSSIM [16] and the proposed method (Fuzzy Color Structural Similarity, FCSS). To assess the match between these measures and the survey perceptual observations, we re-scaled similarity measures results to the interval [1,10].…”

Section: Experimental Studymentioning

confidence: 99%

“…This similarity measure is extended to the Multiscale Structural Similarity Index (MSSIM) in [15]. In turn, in [16], a color comparison criterion is combined with MSSIM. In the approach [17], SSIM scores are weighted by region type.…”

Section: Introductionmentioning

confidence: 99%

Perceptual similarity between color images using fuzzy metrics

Greova

Morillas

2016

Journal of Visual Communication and Image Representation

View full text Add to dashboard Cite

In many applications of the computer vision field measuring the similarity between (color) images is of paramount importance. However, the commonly used pixelwise similarity measures such as Mean Absolute Error, Peak Signal to Noise Ratio, Mean Squared Error or Normalized Color Difference do not match well with perceptual similarity. Recently, it has been proposed a method for gray-scale image similarity that correlates quite well with the perceptual similarity and it has been extended to color images. In this paper we use the basic ideas in this recent work to propose an alternative method based on fuzzy metrics for perceptual color image similarity. Experimental results employing a survey of observations show that the global performance of our proposal is competitive with best state of the art methods and that it shows some advantages in performance for images with low correlation among some image channels.

show abstract

Structural Similarity Measure for Color Images

Cited by 44 publications

References 24 publications

Learning to generate images with perceptual similarity metrics

Learning to generate images with perceptual similarity metrics

Fuzzy averaging filter for impulse noise reduction in colour images with a correction step

Perceptual similarity between color images using fuzzy metrics

Contact Info

Product

Resources

About