A Haar Wavelet-Based Perceptual Similarity Index for Image Quality Assessment (1607.06140v4)

Abstract: In most practical situations, the compression or transmission of images and videos creates distortions that will eventually be perceived by a human observer. Vice versa, image and video restoration techniques, such as inpainting or denoising, aim to enhance the quality of experience of human viewers. Correctly assessing the similarity between an image and an undistorted reference image as subjectively experienced by a human viewer can thus lead to significant improvements in any transmission, compression, or restoration system. This paper introduces the Haar wavelet-based perceptual similarity index (HaarPSI), a novel and computationally inexpensive similarity measure for full reference image quality assessment. The HaarPSI utilizes the coefficients obtained from a Haar wavelet decomposition to assess local similarities between two images, as well as the relative importance of image areas. The consistency of the HaarPSI with the human quality of experience was validated on four large benchmark databases containing thousands of differently distorted images. On these databases, the HaarPSI achieves higher correlations with human opinion scores than state-of-the-art full reference similarity measures like the structural similarity index (SSIM), the feature similarity index (FSIM), and the visual saliency-based index (VSI). Along with the simple computational structure and the short execution time, these experimental results suggest a high applicability of the HaarPSI in real world tasks.

• The paper introduces HaarPSI, a novel metric that leverages Haar wavelet decomposition to emulate human visual perception in image quality assessment.
• It outperforms state-of-the-art metrics like SSIM, FSIM, and VSI in correlation with human opinion scores while reducing computational costs.
• HaarPSI’s efficient design makes it a promising tool for real-time image and video quality evaluation in practical digital applications.

A Haar Wavelet-Based Perceptual Similarity Index for Image Quality Assessment

The paper "A Haar Wavelet-Based Perceptual Similarity Index for Image Quality Assessment" presents a novel method known as the Haar Wavelet-Based Perceptual Similarity Index (HaarPSI) for assessing image quality using full reference (FR) models. The goal of HaarPSI is to mimic human perception in evaluating visual quality, taking into account the distortions introduced during image compression, transmission, or restoration processes. The authors demonstrate that HaarPSI outperforms other state-of-the-art metrics like SSIM, FSIM, and VSI in terms of correlation with human opinion scores across several benchmark datasets.

Methodology and Implementation

The foundation of HaarPSI lies in Haar wavelet decomposition, which captures both high-frequency and low-frequency components of images. Specifically, HaarPSI calculates local image similarities through high-frequency Haar wavelet coefficients and evaluates the significance of different image regions using low-frequency components. This approach seeks to mimic functional properties of the human visual system (HVS), leveraging orientation and spatial frequency selectivity mechanisms.

By adopting the Haar wavelet, known for its computational efficiency, HaarPSI addresses computational cost concerns that often accompany image quality metrics. This simplicity not only reduces computational overhead but, surprisingly, improves correlation with human assessments compared to metrics that use more complex representations.

Experimental Validation and Results

The consistency of HaarPSI with human visual perception was tested on four comprehensive datasets: LIVE, TID2008, TID2013, and CSIQ. These databases feature a wide range of distortion types, including Gaussian noise, JPEG compression artifacts, and blur, among others. HaarPSI consistently achieved higher Spearman Rank Order Correlations (SROCC) with human mean opinion scores than other evaluated FR image quality metrics across these datasets. Notably, only in the case of TID2013's specific test conditions did VSI surpass HaarPSI.

Furthermore, the experiments demonstrated that HaarPSI requires significantly less computational time compared to other high-performing metrics like FSIM and VSI, making it suitable for real-time applications, such as video compression systems, where low latency is critical.

Implications and Future Directions

The results underline the potential practicality of HaarPSI for various real-world applications, including optimizing and monitoring image and video transmission systems. Its performance in detecting distortions common in everyday digital media environments highlights HaarPSI's utility as a robust metric for perceptual quality assessment.

Theoretically, HaarPSI contributes valuable insights into the development of image quality measures by illustrating how basic models of neural function, when correctly implemented, can rival and surpass more complex methods. This discovery prompts further investigation into simple neural-inspired algorithms' potential to mimic human perception.

Future research could explore HaarPSI's application and adaptability to other domains of image processing, possibly incorporating machine learning techniques to extend capabilities to no-reference (NR) scenarios or further refine the weighting mechanisms to enhance specific distortion detection. Integrating divisive normalization processes could also be examined to potentially refine HaarPSI’s perceptual accuracy further.

In conclusion, this paper provides a substantial advancement in image quality assessment, offering an efficient and highly applicable tool for contemporary digital media challenges. By aligning computational efficiency with perceptual relevance, HaarPSI demonstrates that simplicity in design can achieve substantial effectiveness in modeling complex human visual perception phenomena.