Application of electronic techniques in the evaluation of picture quality.
Merian, Richard Frederic
MetadataShow full item record
A study was previously made by Dr. Duncan E. Macdonald and his group at the Optical Research Laboratory of Boston University. This study dealt with the determination of photographic image quality by edge analysis, where an edge here was defined as B greater than or equal to ten percent occurring between a successive trough and peak of a micro-densitometer trace across the photograph. ∆ B is defined as the brightness change across the edge and B is the minimum absolute brightness where this change took place. This thesis is a further study into the work done by Dr. Macdonald. The analysis of his data was done from direct measurements of the paper recorded micro-densitometer traces, but in this study it was proposed that the same or similar data could be recorded automatically using electronic equipment. This thesis presupposes no quantitative definition of an edge, but suggests that with further work a quantitative definition of an edge may be determined. An edge is simply defined as the boundary which separates a picture element from its surroundings on a photograph so that the object is discernible to the eye. We assume that the quality of the picture is a function of its edges and that the edge can be described by two variables; 1) the slope or rate of change of brightness across the edge, and 2) the relative brightness difference across the edge. The equipment used was designed to measure the magnitude of the relative slope, and count the number of slopes of given magnitude in a unidirectional scan across the photograph. Other equipment was designed to measure the magnitude of the reflectivity difference across the edge and count the number of occurences for a given magnitude per linescan. An edge is defined electronically as the change in reflectivity from black to white of a uni-directional scan between any two points where the derivative of this linescan waveform is zero. In this definition no notice is taken as to whether an observer can detect the edge. Five aerial photographs were used in this study graded by impartial observers. Three of the photos are aerial photographs of trees graded as (A) unus~ble, (B) just usuable, (C) excellent. The other two pictures are aerial photos of the Boston suburban area graded as (P) unusable and (G) excellent. These photographs were mounted on a rotating drum which can be adjusted so that any part of the drum can be scanned. The drum was illuminated from a suitable light source. The target is imaged on a pinhole through an optical system. Behind the pinhole is a photo-multiplier tube. The output of the photo-tube serves as the input to a D.C. amplifier. The output of the amplifier is then a voltage scan of a line on the target. To work with the derivative of this waveform the voltage function is then differentiated and level selected. All derivatives above the level selected are counted on an electronic counter. By varying the level selector from zero to maximum level all derivatives are counted and magnitudes recorded. To get the reflectivity difference across an edge a circuit is used which allows an output of the linescan signal to appear only when the derivative is positive. Thus the reflectivity difference across an edge appears relative to a common base line irrespective of what background reflectivity the edge occurred. This in turn was electronically level selected and counted in the same manner as was the derivative. The data consisted of between 1, 500 and 2, 000 counts for every curve drawn. This data was reworked and presented in graphical form, as shown in figures XVII through XX, as cumulative frequency of counts plotted against the magnitude of the relative derivative of the edge in one case and the magnitude of relative reflectivity across an edge in the other. These curves show quite clearly that a difference exists between the good and poor pictures. The excellent photos have slopes that are much greater than the just usuable or unusable pictures. The higher magnitude slopes seem to indicate the quality of the photograph, the lower value slopes being fairly uniform throughout. Similarly the higher reflectivity across an edge appears on the best quality photographs with commensurate degrading for poorer quality pictures. With the knowledge that the edge criterion chosen can lead to objective results for picture quality, it would be interesting to carry on this work in experimentally determining a definition for the threshold photographic edge. This threshold edge is defined as the minimum observable boundary permitting a picture element to be seen in a photograph by a human observer. This could be done by using a large sample from photo-interpreters and the apparatus already perfected for this thesis.
Thesis (M.A.)--Boston University