W11: Is quality subjective or objective? /진신

 I think: Subjective quality vs objective indicators. (references)

Taking photos and videos as examples, the evaluation of photo quality is subdivided into subjectivity and objectivity.

For subjective feelings, we can usually measure the quality of the evaluation index itself based on the accuracy, consistency, stability, and monotony of the prediction. The so-called accuracy refers to the similarity between subjective evaluation scores and objective evaluation index scores; consistency refers to that it should not only perform well for certain types of videos/images but should perform well for all types of videos/images; stable; Sexuality means that the value of each evaluation of the same video/image should be the same or the error should be within the acceptable range; monotonicity means that the evaluation score should show a corresponding single increase or decrease as the MOS score increases or decreases.

From an objective evaluation point of view, we usually use numerical values ​​to classify evaluation indicators based on errors, evaluation indicators based on perception models and image structure information, and evaluation indicators based on machine learning. Based on the error evaluation index, the compressed image is compared with the original image, and the difference (called noise or error) between the two images is calculated. Representative indicators are Mean square error (MSE) and Peak signal noise ratio (PSNR). Based on the perception model and image structure evaluation index, the human visual system (HVS) is introduced to transform the image quality degradation into changes in perceptual structure information and changes in some perceptual phenomena (brightness, contrast, viewing distance), and mathematically how humans perceive these errors Modeling, Structure similarity Index (SSIM), Just Noticeable Difference (JND). Evaluation indicators based on machine learning are commonly used to measure long-term videos. Starting from a certain trainable model, the results of error-based evaluation indicators or perceptual evaluation indicators are compared with subjective MOS score, and the model is fine-tuned to make it follow suit. An improvement over time, or the fusion of multiple evaluation index systems. These multiple indicators include indicators that measure image quality and indicators that measure time quality. The typical evaluation indicators based on machine learning are Visual Multimethod Assessment Fusion(VMAF).

Combining the above two methods of evaluation, I believe that quality is subjective, while the objective one should be known as an indicator. What is your view on this?

Comments

Post a Comment

Popular posts from this blog

W4: What are the pros and cons of using Naver vs Google for research purposes?

Google Scholar: In terms of academic research, Google Scholar is a web search engine that can search academic articles for free. Its source of information is also professional and rich. For example, articles from academic publishers, etc. There are also most of the published academic journals in the world. It also supports advanced search functions such as keywords, classification and filtering, and you can find the most relevant research documents in the entire academic field in a few seconds. Every search result on Google Scholar will show citations. This provides a good extended search clue, which is convenient for users to directly view related papers. As a Chinese student, even in mainland China, Google Scholar cannot directly access and use it. We will still use VPN and other methods for reference when conducting research. NAVER: And naver is more like an interactive community. There are more academic websites in Korea.  As a Chinese student, I think Naver rarely uses for acade
Read more

W2 : After watching Mitch Resnick’s TED video

 Through this video, I realized that programming education is not only as deep and professional as I know it, but also more flat and interesting. With the change of the times and the advancement of technology, the times we experience are indeed constantly changing. I believe that programming is just a way of learning like reading. It does not depend on any subjective intentions but exercises people’s logical thinking and creativity. China There is a common saying that “giving others fish is better than fishing for others.” I believe that programming does not exist as a short-lived technology, but behind it is that people need to keep learning with the times and have an open attitude. The purpose of early childhood education is different from the professional education we receive now. Early childhood education is more inclined to teach children the way of thinking, rather than the part of applying knowledge or technology in a professional subject as we are now learning. Therefore, addin
Read more

W13/After reading “Categorization”/HARI KIM

Summary) Categorization is a kind of list for dividing topics, and it helps users find information easily. All documents must be in at least one category. Therefore, categories must be well chosen. How are the numerous articles on Wikipedia categorized? First, the category should be chosen as the most helpful topic when someone reads the article. And usually an article doesn't exist in both the top and the subcategories at the same time. If there are few documents in the category you have in mind, you should check to see if they belong to a subcategory. It is also important to maintain a "neutral" perspective when categorizing. New/Interesting things I learned) I learned that usually one document does not exist in both the upper and subcategories at the same time. Before I knew this information, I had made a mistake in categorizing documents. In categorizing articles about rivers, the article was already part of a subcategory, but I have made mistakes by adding a top cate
Read more