I suspect that any advantage has to do with image composition and image intent. That is to say, the histogram of the image, and the output device. Any advantage of one method over the other would, I assume, lie in the "eye" of the editor.
As far as file size, first, I would only judge size based on the final saved file on disk, but that’s a personal preference. I haven’t read the article, but Blatner and Fraser may have been referring to a smaller file size once flattened. It’s just a guess since, as I say, I haven’t read what you are referring to.
That’s my two cents anyhow.
Peace,
Tony
JJ,
The two methods are quite different but may lead to similar results for SOME images. The differences should be least noticeable in the highlights, more noticeable in the low mid-range, and most noticeable in the shadow areas.
The Levels middle slider adjusts the gamma, bowing the input-output curve upwards or downwards, depending on the direction in which you move the slider, keeping the ends of the curve anchored.
The screen blend mode anchors the upper right corner of the input-output curve. while sliding the lower left corner vertically upwards, maintaining a straight line input-output relationship. At 100% opacity the effect is maximum. Lowering opacity from there simply mutes the effect, i.e., slides the lower left corner back down in proportion to the opacity reduction.
George
JJ
Sorry, I was too quick on the trigger. My remarks apply when screening with a constant-valued layer on top of the image. For a self-screen, in which the screening layer is identical to the underlying imasge, both high and low ends of the input-output curve are anchored and you do get an upward bulge in the curve resembling what you get by moving the Levels slider left.
George
Taking a closer look at a screen mode self blend vs moving the middle slider in Levels:
The input-output curve for Levels bows upward as the middle slider value is increased above 1, anchoring 0 and 255 values and increasing all values in between following the "bow" line. The curve this produces for output is input raised to the rcciprocal of the slider value (a "power" curve). If the slider value is 2, for example, output equals input to the 1/2 power, i.e., the square root of the input. Keep in mind that input and output need to be "normalized" by dividing by 255, or else transformed from a 0-255 range to a 0-1 range. The root of a decimal number being greater than the number, the output is thus greater than the input for all fractional powers. A middle slider value of 2 gives its reciprocal (1/2) as the power. and so all outputs are greater than the inputs and the image brightens all over, the brightening being greatest at middle inputs and tapering to no change at the extreme ends.
The screen blend mode for a self blend also brightens all inputs, but not on a strict power curve. There is an S-like warp in the curve it produces and the S may cross the power curve. Comparing the self-screen curve with the square root curve cited above, the crossover (point where the two resultant output values are equal) happens to occur at a normalized input of .382 (color value 97). Above that input the screened value is slightly higher than the Levels value (again for the specific middle slider value of 2) and the screened output value is slightly lower than the Levels output value for inputs less than .382 (97). For other middle slider values (gamma corrections) the crossover will move to a different point.
The self-blend screen curve relating the color channel value at any pixel site to its new value after self-screening is an unchanging curve. You can cunstruct any number of gamma correction power curves by changing the Levels middle slider position until you get one that matches fairly well with the self-screen curve, but there will never be an exact match. A gamma of 2 comes pretty close. Were this a classroom, a good exercise for the student would be to find the best match gamma correction value.
George