i74 



NATURE 



[June 21, 1888 



frequently chafed under the sense of inability to ver- 

 bally explain hereditary resemblances and types of 

 features, and to describe irregular outlines of many 

 different kinds, which I will not now particularize. At 

 last I. tried to relieve myself as far as might be from 

 this embarrasment, and took considerable trouble, and 

 made many experiments. The net result is that while 

 there appear to be many ways of approximately effecting 

 what is wanted, it is difficult as yet to select the best of 

 them with enough assurance to justify a plunge into a 

 rather serious undertaking. According to the French 

 proverb, the better has thus far proved an enemy to the 

 passably good, so I cannot go much into detail at present, 

 but will chiefly dwell on general principles. 



Measure of Resemblance. — We recognize different 

 degrees of likeness and unlikeness, though I am not 

 aware that attempts have been as yet made to measure 

 them. This can be done if we take for our unit the least 

 discernible difference. The application of this principle to 

 irregular contours is particularly easy. Fig. 1 shows two 

 such contours, A and B, which might be meteorological, 

 geographical, or anything else. They are drawn with firm 

 lines, but of different strengths for the sake of distinction. 

 They contain the same area, and are so superimposed as to 

 lie as fairly one over the other as may be. Now draw a 

 broken contour which we will call C, equally subdividing 

 the intervals between A and B ; then C will be more like 

 A than B was. Again draw a dotted contour, D, equally 

 subdividing the intervals between C and A ; the likeness of 

 D to A will be again closer. Continue to act on the same 



Fig. 



principle untila stageis reached when the contour last drawn 

 is undistinguishable from A. Suppose it to be the fourth 

 stage ; then as 2 4 = 16, there are 16 grades of least-discern- 

 ible differences between A and B. If one of the contours 

 differs greatly in a single or few respects from the other, 

 reservation may be made of those peculiarities. Thus, if 

 A has a deep notch in its lower right-hand border, we 

 might either state that fact, and say that in other respects 

 it differed from B by only 16 grades of unlikeness, or we 

 might make no reservation, and continue subdividing 

 until all trace of the notch was smoothed away. It is 

 purely a matter of convenience which course should be 

 adopted in any given case. The measurement of resem- 

 blance by units of least-discernible differences is applicable 

 \o shades, colours, sounds, tastes, and to sense-indications 

 generally. There is no such thing as infinite unlikeness. 

 A point as perceived by the sense of sight is not a 

 mathematical point, but an object so small that its 

 shape ceases to be discernible. Mathematically, it 

 requires an infinitude of points to make a short line ; 

 sensibly, it requires a finite and not a large number of 

 what the vision reckons as points, to do so. If from thirty 

 to forty points were dotted in a row across the disk of the 

 moon, they would appear to the naked eyes of most 

 persons as a continous line. 



Description within Specified Limits. — It is impossible to 

 verbally define an irregular contour with such precision 

 that a drawing made from the description shall be undis- 

 tinguishable from the original, but we may be content with 

 a lower achievement. Much would be gained if we could 



refer to a standard collection of contours drawn with 

 double lines, and say that the contour in question falls 

 between the double lines of the contour catalogued as 

 number so-and-so. This would at least tell us that none 

 of the very many contours that fell outside the specified 

 limits could be the one to which the description applied. 

 It is an approximate and a negative method of identifica- 

 tion. Suppose the contour to be a profile, and for sim- 

 plicity's sake let us suppose it to be only the portion of 

 a profile that lies below the notch that separates the brow 

 from the nose and above the parting between the lips, and 

 such as is afforded by a shadow sharply cast upon the wall 

 by a single source of light, such as is excellently seen when 

 a person stands side- ways between the electric lantern and 

 the screen in a lecture-room. All human profiles of this 

 kind, when they have been reduced to a uniform vertical 



Fig. 2. 



scale, fall within a small space. I have taken those given 

 by Lavater, which are in many cases of extreme shapes, 

 and have added others of English faces, and find that 

 they all fall within the space shown in Fig. 2. The 

 outer and inner limits of the space are of course not 

 the profiles of any real faces, but the .limits of many 

 profiles, some of which are exceptional at one point and 

 others at another. We can classify the great majority 

 of profiles so that the whole of each class shall be in- 

 cluded between the double borders of one, two, or some 

 small number of standard portraits such as Fig. 3. I am 

 as yet unprepared to say how near together the double 

 borders of such standard portraits should be ; in other 

 words, what is the smallest number of grades of un- 

 likeness that we can satisfactorily deal with. The process 

 of sorting profiles into their proper classes and of gradually 



Fig. 3. 



building up a well-selected standard collection, is a 

 laborious undertaking if attempted by any obvious way, 

 but I believe it can be effected with comparative ease on 

 the basis of measurements, as will be explained later on, 

 and by an apparatus that will be described. 



Classification of Sets of Measures. — Prisoners are now 

 identified in France by the measures of their heads and 

 limbs, the set of measures of each suspected person being 

 compared with the sets that severally refer to each of many 

 thousands of convicts. This idea, and the practical appli- 

 cation of it, is due to M. Alphonse Bertillon. The actual 

 method by which this is done is not all that could be 

 theoretically desired, but it is said to be effective in action, 

 and enables the authorities quickly to assure themselves 

 whether the suspected person is or is not an old malefactor. 

 The primary measures in the classification are four — 



