Confirmation bias, the tendency to interpret information in a way that aligns with one’s preconceptions, can profoundly impact scientific research, leading to conclusions that reflect the researcher’s hypotheses even when the observational data do not support them. This issue is especially critical in scientific fields involving highly noisy observations, such as cryo-electron microscopy. This study investigates confirmation bias in Gaussian mixture models. We consider the following experiment: A team of scientists assumes they are analyzing data drawn from a Gaussian mixture model with known signals (hypotheses) as centroids. However, in reality, the observations consist entirely of noise without any informative structure. The researchers use a single iteration of theK-means or expectation-maximization algorithms, two popular algorithms to estimate the centroids. Despite the observations being pure noise, we show that these algorithms yield biased estimates that resemble the initial hypotheses, contradicting the unbiased expectation that averaging these noise observations would converge to zero. Namely, the algorithms generate estimates that mirror the postulated model, although the hypotheses (the presumed centroids of the Gaussian mixture) are not evident in the observations. Specifically, among other results, we prove a positive correlation between the estimates produced by the algorithms and the corresponding hypotheses. We also derive explicit closed-form expressions of the estimates for a finite and infinite number of hypotheses. Furthermore, we provide theoretical and empirical results for multi-iterationK-means and expectation-maximization, showing that the bias is persistent even after hundreds of iterations of these algorithms. This study underscores the risks of confirmation bias in low signal-to-noise environments, provides insights into potential pitfalls in scientific methodologies, and highlights the importance of prudent data interpretation.

Reduced-reference image quality measures aim to estimate the visual quality of a distorted image with only partial information about the “perfect quality” reference image. In this paper, we present a reduced-reference image quality assessment (IQA) metric based on DCT Subbands Similarity (RR-DSS). According to the assumption that human visual perception is adapted for extracting structural information, the proposed technique measures change in structural information in subbands in the discrete cosine transform (DCT) domain and weights the quality estimates for these subbands. RR-DSS is simple to implement, incurs low computational complexity, and has a flexible tradeoff between the amount of side information and image quality estimation accuracy. RR-DSS was tested with public image databases and shows excellent correlation with human judgments of quality. It outperforms state-of-the-art RR IQA techniques and even several FR IQA techniques.

Measuring image quality becomes increasingly important due to the many applications involving digital imaging and communication. Image quality assessment aims to develop a visual quality metric that correlates well with human visual perception. In this paper, we present a full-reference image quality assessment technique based on DCT Subbands Similarity (DSS). The proposed technique exploits important characteristics of human visual perception by measuring change in structural information in subbands in the discrete cosine transform (DCT) domain and weighting the quality estimates for these subbands. The proposed technique was tested with public image datasets and shows higher correlation with subjective results than state-of-the-art techniques. Another advantage of the proposed technique is its low computational cost.