Abstract

The basic question posed by this study is - WHAT IS THE SENSITIVITY OF PICTORIAL SUBJECTIVE QUALITY TO A GRANULARITY INCREASE? This question is decidedly different than the typically posed question of granularity studies, that is - WHAT IS THE SENSITIVITY OF PERCEIVED GRANULARITY (graininess) TO A GRANULARITY INCREASE? In the subjective evaluations herein, the former question was considered most pertinent to response from the general public, who in toto, do not interpret graininess as an attribute per se, but only that ubiquitous factor known as "quality" - poor, indifferent, or good. All references in this paper to granularity/quality thresholds (i.e., Just Noticeable Difference (JND)) should be considered from this standpoint. This interpretation could be critical in that thresholds for perceived granularity and quality may be at best functionally related in some complex manner. This could occur since quality is a factor which could be dependent on several inter related parameters including graininess. The study result suggests that the above two questions or criterion are at least mutually commensurable in a gross sense. The apparent differences were discerned in the JND's emulated in the digital simulation and those published for conventional pictorial results. General visual results were excellent in terms of comparability to graininess expected from "real" systems. The results also suggest that the subjective evaluation of quality may be scene dependent. The lower complexity (low degree of textural heterogeneity) scene subjective quality varied quite linearly with granularity while the higher complexity (high degree of textural heterogeneity) scene was apparently non-linear, or logarithmic over the same granularity range - the sensitivity for lower granularities higher (smaller JND) than at higher granularities. The general log model was found to correlate reasonably well with low and high complexity scenes. The result is shown to support the concept of information content as one model for subjective quality which is expanded on in Section 5.0 and Appendix B. Experimental data indicated that S - - 6.3 + 2.6 In 0/an) where S represents subjective quality characterized by statistical scaling and an is the Root-Mean-Square (RMS) density or granularity. This equation, over the range of variables considered and statistical confidence attained, reasonably models results for all scene types. However, the low complexity scene subjective quality was better described by the simpler linear expression or S = + 2.6 - 27.1 (aD). The question of the dependency of subjective quality on scene type can only be adequately resolved with further experiments. For now, hypothetical concepts are explored as with the information content model of subjective quality in Appendix B. Three spatial frequency "bandpass filters" for low, medium', and high degree of "busyness" subregions (within the pictorial format) are assumed to operate independently in attempting an explanation of the possible scene dependency response. Past studies suggest that granularity perception sensitivity is maximal for the simpler scenes (e.g., uniformly exposed and developed emulsions). This study shows results, for the type of scenes treated herein, which differ in that at lower granularity levels, the sensitivity of subjective quality to changes in granularity is greatest for higher complexity scenes. The model of Appendix B is also used to supply a rationale for this condition using the bandpass concept as noted above.

Library of Congress Subject Headings

Images, Photographic; Photographic sensitometry

Publication Date

11-1-1981

Document Type

Thesis

Department, Program, or Center

School of Photographic Arts and Sciences (CIAS)

Advisor

Francis, Ronald

Advisor/Committee Member

Granger, Edward

Advisor/Committee Member

Higgins, George

Comments

Note: imported from RIT’s Digital Media Library running on DSpace to RIT Scholar Works. Physical copy available through RIT's The Wallace Library at: TR280.L57 1981

Campus

RIT – Main Campus

Share

COinS