29 June 2018

The Limits of Image Recognition

For the last two posts in this series on image recognition, I used an image which was topical, but little more, to try two Google services.

Shown below is an image I saved from eBay earlier this year. The eBay description said,

FRANK MARSHALL SIGNED LIMITED EDITION MY FIFTY YEARS OF CHESS • Hardcover. 8vo. Horowitz & Harkness. 1942. Vi, 242 pgs. First Edition/First Printing. Signed by Frank Marshall on the limitation page. #2 of a signed limited edition of 500. Illustrated with black and white photos. No DJ. Bound in red cloth with gilt titles. Light rub present to the spine. Previous owner's bookplate on the reverse of the front board. Text is clean and free of marks, binding tight and solid, boards clean with no wear present. The autobiography of one the great American tournament chess players who would later on lend his name to the Marshall Chess Club in Manhattan.”

The main reason I had saved it was because of a pair of posts from last year:-

That first post included a quote from an article by T.A. Dunst:-

When Frank Marshall in 1942 wrote My Fifty Years of Chess, summing up an international chess career and 27 years possession of the United States chess championship, the book contained a great deal more of Thomas Emery than the preface which he supplied.

Could image recognition tell me anything new about Emery's preface?

I fed the top half of the image, the photograph, to images.google.com. It told me,

No other sizes of this image found.
Best guess for this image: photograph

It also suggested 'Visually similar images', all of which were black and white photos, and none of which had anything to do with chess. Then I fed the entire image, a full page of the book, to cloud.google.com/vision. It told me 'Labels':-

93% Text
70% Font
69% History
54% Black And White

It also told me 'Web Entities':-

0.6106 Font
0.5742 White

In other words, it didn't tell me anything I didn't already know. What about the 'Document' analysis on the text portion of the page? It gave me a decent equivalent of OCR analysis, but the output was in Unicode format. Since this would have taken me some effort to convert, I ran the text portion of the image through my normal OCR conversion service, which gave me the following:-

The world needs no introduction to Frank James Marshall, our great American Chess Master, who has played all over the world and achieved successes unparalleled in history.

What the world may not know, however, is that our Frank would probably have done even better in chess tournaments if he had not always been so eager to play for a win. An artist of the chess board, he always played to win so long as there was the remotest possibility of succeeding. As a result, he lost many a game which he could have drawn if he had not set out with that idea in mind.

It has been my privilege to have known Frank over a period of twenty years and to have worked with him during that time. No finer sportsman than Frank Marshall ever sat down to a chess board, nor could anybody hope to have a better friend.

Of this one can be certain: Marshall's name will live to eternity. To paraphrase Shakespeare, "Age cannot wither him, nor custom stale his infinite variety."

New York, 1942.

In fact, I had to correct a few conversion errors in the above text, passages which were rendered perfectly by the Google service. I fed Emery's first sentence to the normal Google search and it gave me a link to a digital copy of the book in Google books. That means Google couldn't match the photo in the book, but it could match the text. This post might not have much to do with image recognition, but it at least has something to do with Marshall and Emery.

No comments: