Best Image

Deterministically choose the best image from an array of images.

Summary

Goal of this project is to abstract the logic and mechanism of programatically chosing the most appropriate image given a target and an array of images by comparing the properties of one image against the properties of a following image within an array. The winner will then be selected to be compare to the next image within the array until all images of the array have been iterated over.

Concepts

This is an early project so ideas may change and things may be renamed, but the core concepts should be reasonably stable.

Target

A "target" is a set of attributes that describe what the best image would be for a given use case. Currently "width" and "height", but in the future it could be anything such as the target luminance, color average, or any other property that could be described in terms of best case / worst case percentages.

MatchBy

A "matchBy" is a collection of "matchers" that can return the averaged score from the percentage score returned from each "matcher". The idea here is to create a score card for different criteria such as the "best-thumbnail" image, "best-banner" image, or "best-veritcal" image.

Matcher

A "matcher" is a method of providing a score of a single attribute that can be described in terms of best-case match (100%) and worst-case match (0%.) The idea is to focus on the method of determining of a single property and normalizing the output as a percentage value.

Matchers

A list of provided matchers:

• orientationMatch
• ratioMatch
• scaleToContainMatch
• scaleToCoverMatch

orientationMatch

ratioMatch

scaleToContainMatch

scaleToCoverMatch

Caveats

This algorithm chooses the "best" image from a list of images. The first image in the array is assumed to be the de facto "best" image. It may be surprising if an impractical image is selected in some scenarios.

Array order is important

This algorithm assumes that the best image must be "better" than the previously selected image. If the images being compared have the "same" score then the previous selection shall remain the "best" image.

Best of the worst

This algorithm assumes there only valid images in the array to select from. If an obviously poor image has been selected with a 0% match and is compared to another image whose score is not better than a 0% match then the "best" image may appear to be a silly and arbitrary selection. To prevent these kinds of surprises, the array of images should be pre-filtered to remove any unqualified choices.