Bertillon Chart

[Colette]

Thechart depicts the method of taking thirteen measurements for identification using the Bertillon method. An American eagle is shown at the top and a shield with stars and stripes at the bottom, and decorative scroll work is used in each of the four corners. Grand Rapids Police Department Guide for Use of the Alphonse Bertillon Identification System is printed at the bottom of the chart.

Now, some of the earliest mug shots ever taken are on display at The Metropolitan Museum of Art in New York City. The black-and-white photos were once on the cutting edge of how police identified suspects.

Police stations would hang charts showing rows of eyes, ears, noses and other body parts to help officers identify suspects using Bertillon's system. Courtesy of The Metropolitan Museum of Art, New York hide caption

"Take 10 different people, take pictures of their ears and you'll be to identify each and every one of them because we all have different facets to our ears. Some of us have longer earlobes, some shorter, some thicker, some thinner," he says.

It's an observation that Bertillon championed back in the late 1800s. He went on to create a system for identification that the Paris Police Prefecture adopted in 1882, giving rise to the modern mug shot.

"Men can grow facial hair to cover their chin, but you can't change the shape of your nose except through surgery. And you can't change the contours of your ear," explains Mia Fineman, one of the curators for the Met's new photo exhibition Crime Stories: Photography and Foul Play, which features some of Bertillon's work.

French criminologist Alphonse Bertillon's (left) techniques for identifying criminals in the late 19th century set the template that police use today. Mondadori Portfolio via Getty Images hide caption

"Whenever you go through an airport or at a train station and anything else and somebody asks to see your identification document, that all has roots in the late 1800s and the work of people like Bertillon and his contemporaries," Finn says.

Many historical visualizations capture important statistics of historical events, and therefore are of great interest to scholars in humanities and social sciences. In numerous cases, the original datasets are lost. It is desirable to reconstruct the datasets from the visualization images. Conventionally, reconstructing such datasets requires manually measuring the visual attributes.

MI3 is a workflow and system for data reconstruction from historical visualization images. MI3 turns the task of data reconstruction from historical visualizations into interactive classification tasks. It uses machine-initiated intelligent interaction that features algorithmic sampling and default labeling to reduce user efforts in interactive classification tasks.

Data reconstruction tasks can be converted to interactive classification tasks. In the figure above, we show an example of turning discrete bar chart data reconstruction into a sequence of three tasks: block detection, block grouping, and root detection. Each of the three tasks can be formulated as a classification task.

Block detection extracts the rectangle glyphs from the image. Block detection can be formulated as algorithmically proposing a collection of visual objects in the image and classifying them as blocks or non-block.

Block grouping groups the detected blocks into discrete bars. Block grouping can be formulated as conceptually building a graph with blocks, with edges denoting blocks that belong to the same discrete bar, and classifying edges as should be connected or not.

Root detection decodes the position information attached to each discrete bar. Root detection can be formulated as proposing feature points in discrete bars, and classifying them as true detections or not.

Such task conversions typically require preprocessing by image processing algorithms. MI3 consists of a collection of image processing algorithms to accommodate the need for detecting different types of visual objects.

Algorithmic sampling selects data objects according to a predefined metric that predicts the potential benefit of labeling a data object to the learning process. Active learning methods, such as uncertainty-based sampling, can be used for algorithmic sampling. In general, any algorithm that can provide an ordering for the data objects may serve the purpose.

Default labeling assigns a default label to each data object when data objects are presented in a user interface for dynamic labeling. Default labeling is supported by models that incrementally learn from the user's input. In this way, the user's input's value is maximized, as it can potentially affect the labeling result for all the unlabeled data objects. If the system can predict some labels correctly, this can reduce the number of interactions that a human user has to perform in labeling the data objects.

There were few graphical innovations, and, by the mid-1930s, the enthusiasm for visualization which characterized the late 1800s had been supplanted by the rise of quantification and formal, often statistical, models in the social sciences. Numbers, parameter estimates, and, especially, standard errors were precise. Pictures were- well, just pictures: pretty or evocative, perhaps, but incapable of stating a "fact" to three or more decimals. Or so it seemed to statisticians.

In this period graphical methods were used, perhaps for the first time, to provide new insights, discoveries, and theories in astronomy, physics, biology, and other sciences. As well, experimental comparisons of the efficacy of various graphics forms were begun, e.g., (Eells:1926), and a number of practical aids to graphing were developed. In the latter part of this period, new ideas and methods for multi-dimensional data in statistics and psychology would provide the impetus to look beyond the 2D plane.

Du Bois portrait

At the Paris Exposition in 1900, W. E. B. Du Bois compiled an exhibit of hundreds of graphs and photographs depicting the history of Negroes in America, including over 60 statistical charts, graphs and maps.

Exports from England and Ireland

In one of the first statistical textbooks, Arthur Bowley (1901) illustrated an arithmetic and graphical analysis of time-series data using the total value of British and Irish exports from 1855-1899. He presented a line graph of the time-series data, supplemented by overlaid line graphs of 3-, 5- and 10-year moving averages. His goal was to show that while the initial series showed wide variability, moving averages made the series progressively smoother.

On Chart Uniformity (pg. 1) - English Translation

On Chart Uniformity (pg. 2) - English Translation

On Chart Uniformity (pg. 3) - English Translation

On Chart Uniformity (pg. 4) - English Translation

On Chart Uniformity (pg. 5) - English Translation

On Chart Uniformity (pg. 6) - English Translation

This work by Cheysson and Fontaine, reported by Bertillon, was the first proposal of standards for graphical presentation. The paper "Proposals to bring uniformity in the preparation of charts" discusses some key recommendations, including cautious use of symbols and hieroglyphs, and sparing use of comparison by areas.

It is proposed that x and y scales be constructed so that the average behaviour corresponds to a curve of 45 degrees. Other attempts to formulate standards for graphical procedures at the International Statistical Congress are discussed in detail.

Maunder's butterfly diagram

1904 Butterfly Diagram: Sunspots

Use of the "butterfly diagram'' to study the variation of sunspots over time, leading to the discovery that they were markedly reduced in frequency from 1645--1715 (the "Maunder minimum''). [Earlier work, started in 1843 by H. Schwabe, showed that sunspots exhibit an approximately twenty-two year cycle, with each eleven-year cycle of sunspots followed by a reversal of the direction of the sun's magnetic field]

Rosele portrait

Trellis-like time series graphs of tuberculosis

Trellis-like time series graphs of infant mortality

3D Histogram: The course of death in Saxony

First International Hygiene-Exhibition in Dresden, with 259 graphical-statistical figures of 35 national and international exhibitors and more than 5 million visitors. [Roesle also wrote publications which dealt with the structure of graphical-statistical displays citeRoesle:1913.]

Hertzprung's first 1911 graphs

early Hertzsprung-Russell diagram

modern Hertzsprung-Russell diagram

The Hertzsprung-Russell diagram, a log-log plot of luminosity as a function of temperature for stars, used to explain the changes as a star evolves. It provided an entirely new way to look at stars, and laid the groundwork for modern stellar physics and evolution, developed independently by

Photograph of the Parade of Statistical Graphics

Parade of statistical graphics, May 17, 1913, including large graphs on horse-drawn floats, and a photograph with people arranged in a bell-shaped curve

Moseley portrait

Moseley graph image

Discovery of the concept of atomic number, based largely on graphical analysis (a plot of serial numbers of the elements vs. square root of frequencies from X-ray spectra) The linear relations showed that the periodic table was explained by atomic number rather than, as had been supposed, atomic weight, and predicted the existence of several yet-undiscovered elements

Correspondence course in graphical methods (20 lessons for $50, supplemented by a book of 100 specimen illustrations of bar, curve, and circle diagrams; entended title includes "There's an idea in every chart'')

de Martonne portrait

Distribution of nationalities in the the countries dominated by Roumanians (2925 x 1959; 1026K)

Use of ethnographic maps, showing the distribution of mixed nationalities, played an important role in redrawing national boundaries of Central Europe and the Balkans following World War I

3a8082e126