FR2731535A1 - Reference mark and method of locating information from this mark - Google Patents

Abstract

The reference mark is placed in a fixed position relative to the text to be read. The characters forming the words are of the reference mark have an order and height ensuring the mark spans the space of the text to be read. The space between each word of the mark is greater than the spaces between the words of the text outside the area to be read. The height of the words in the mark is greater than that of words outside the area. The number of words is greater than the number of consecutive words that form a line of the words outside the area. The surface to be read is scanned and the image filtered to locate the reference mark, using row and column filtering initially then filtering according to length. This is followed by column filtering to extract words by height.

Description

LOCATION MARK
AND METHOD FOR LOCATING INFORMATION
BY ADDING THIS BRAND
The invention relates to a registration mark intended to be added to information which it is sought to locate on a complex image, comprising for example any texts and graphics. It also relates to a method of locating this information, which can be an address, a digital code or a logo for example.

This invention is of particular interest in industrial applications such as reading the address on postal objects, finding a number on spare parts, automatic reading of license plates on road images, reading of 'labels on supermarket items or trolleys in an industrial environment.

In these applications, the difficulty in locating information comes from the fact that it is located on an image already loaded with information of all kinds.

Thus, concerning the location of postal addresses on a flat face of objects to be sent, such as envelopes, packages or magazines wrapped in transparent plastic film, which is a fundamental step in current automatic mail sorting processes, it is made difficult because of the texts and advertising graphics that surround it. This leads to false address detections or the detection of erroneous lines in the vicinity of this address.

The French patent application published under the number 2 657 982 relates to a method of locating an address on mail to be sorted, consisting in associating with the address a specific address mark, different from almost all the characters and graphics on the mail, and detect it in the digital image of the mail. This method has the drawback on the one hand of using a mark which is difficult to print at the same time as the address by a conventional printing method and on the other hand of requiring powerful tracking software, costly in time.

The object of the invention is to overcome the complexity of the flat object on which the information to be read is located, by affixing a mark in a determined position with respect to this information which makes it possible to locate it easily with a low error rate.

For this, a first object of the invention is a marker for information with which it is associated, in a complex image comprising words consisting of letters and numbers in block letters and graphics, the relative positions of the mark and of the information being fixed, characterized in that it is constituted by a line of several words such as - the characters constituting the words of the mark have a font and a size intended to allow blocking of the words; - the space between each word of the mark is determined, greater than the spaces between the words of the image using the same characters as the words of the mark and greater than the blocking threshold; - the size of the words is different from the size of most of the words in the image which are consecutive and likely to form a line on the image; - the number of words is greater than the number of consecutive words likely to form a line on the image and of sizes identical to that of the words of the mark.

According to another characteristic, the mark consists of a line of words of identical size, at least four in number.

Another object of the invention is a method of locating information on a two-dimensional image by adding a mark defined by the invention, comprising a phase of locating the mark from the binary image of the original image followed by a location of the information from this mark knowing their relative positions, characterized in that the marking of the mark comprises a first phase of filtering the binary image, defined by lines and columns, to keep only the words likely to belong to the mark, consisting of - a first step of blocking the words line after line, delivering segments of black dots leading to a single related component per word; a second step of filtering the segments along their length, line after line, intended to eliminate the words of length less than the length of a word of the mark and the words of length greater than the total length of the mark; - a third step of filtering the segments on the height, column by column, intended to keep the words of height close to that of the words of the mark; and a second phase of localization of the mark on the image consisting of - a first step of labeling the related components obtained in the previous phase; a second step of filtering the related components over their length and their height, intended to keep the words whose size is compatible with that of the words of the mark; - a third step of merging the components located at the end of the previous step, online; - a fourth step of filtering the lines of merged components, on their dimensions, height and length, and on their number of words.

Other characteristics and advantages of the invention will appear on reading the following description of a particular example of application, this description being made in relation to the attached drawings in which - Figures la and lb represent some examples of marks respectively according to the invention and not responding to the invention; - Figure 2 is the original binary image on which we seek to locate information - a postal address - Figure 3a is an image showing segments obtained after blocking line by line; - Figure 3b is an image showing the segments obtained after filtering, column after column, on the height of the segments; - Figure 4 is the image obtained after plugging and filtering the components; - Figure 5 is the image obtained after labeling and filtering its components on their two dimensions; - Figures 6a to 6e show the different criteria to be met between two consecutive components to allow their online fusion; - Figure 7 is the image on which are identified the lines formed after merging the words; - Figure 8 is the image on which is identified the selected mark; - Figure 9 is the image on which the information is located from the mark.

The elements bearing the same references in the different figures fulfill the same functions for the same results.

Given an image, composed mainly of words in letters or numbers, in block letters, and graphics, carrying information to be located, the registration mark with which it is associated must be easily locatable and distinct from everything graphics may appear on the image. For this, the brand must have characteristics rarely encountered in an image containing ordinary texts. In addition, the mark must be able to be easily printed by any conventional printer, preferably at the same time as the information to be located.

For this, the mark is constituted by a line of several words which must have the following characteristics - the characters constituting the words of the mark and produced by black dots on a background of white dots, have a font and a size such as the words are pluggable, in other words the white horizontal areas inside the words are less than a threshold characteristic of a plugging operation consisting in blackening the white areas between the black dots constituting the words; - the spaces between the words are greater than the spaces between the words of the image using the same characters and greater than the characteristic threshold of the blocking previously mentioned; the size of the words of the mark is chosen to be different from the size of most of the words which are placed consecutively and capable of forming a line on the image, and preferably identical for all the words of the mark; - the number of words making up the mark is chosen large enough to reduce the risk of detecting it in other places in the image, therefore greater than the number of consecutive words likely to form a line on the image and which have sizes identical to those of the brand's words. In case the words of the brand are the same size, the number is at least equal to four.

Thus the total length of the mark is known and determined by the number of words, their size and their spacing, so that the risk of locating words of the image satisfying all the constraints previously stated is very low.

During the localization of the information, the characters constituting the words of this registration mark are not read and several configurations of words meeting the previous criteria can be used concurrently. The figure gives some examples of marks according to the invention, consisting of at least four words in block letters, these words can be formed of numbers or letters, different or identical, of a combination of block characters or signs leading to words of the same size, while the examples given in FIG. 1b do not constitute marks according to the invention, being made up of words whose characters have a variable height creating white horizontal ranges that are too large, too poorly spaced, too short, variable in length, or too short.

The examples in FIG. 1 a represent the marks according to the invention arranged horizontally, ie parallel to a dimension of the image, but are not limiting, the marks being able to consist of a column of words arranged vertically for example.

As for the position of the brand in relation to the information to be located, it must be determined according to the nature of this information. Thus, in the case of a postal address to be located on a mail, we choose to place the mark below the last line of the address, that is to say under the routing line, comprising the postal code and city, because the number of lines of an address is variable. In all cases, the mark must be spaced from the components of the image on its four sides in order to avoid any fusion with parasites when locating the mark.

The mark defined according to the invention has the following advantages compared to the mark of the prior art: it can be printed by any current conventional printer and its search is made using a simple process, of rapid execution while being robust and inexpensive in machine time.

The second object of the invention is a method of locating information on an image which consists first of all in locating a mark, defined as the first object of the invention, from a binary image 1 of the original two-dimensional image, as shown in FIG. 2, an image composed of words in letters 2 or in numbers 3 in block letters, and graphics 4, then in locating the information from this mark, knowing their relative positions. The binary image is made up of black dots and white dots arranged in lines and columns, perpendicular to the lines, as shown in the following figures 3a and 3b
The method according to the invention comprises two phases, the first consisting in filtering the binary image in order to keep only words likely to belong to the mark, and the second phase consisting in labeling and grouping words in lines then in locating the marking line.

The first phase of the method, which is a filtering phase of the binary image, comprises a first step of blocking the words in order to obtain a single connected component per word. As shown in FIG. 3a, the blocking of a word 6, line by line, consists in blackening white dots 7 located between two black dots 8 on the same line 9 parallel to one of the dimensions of the image, horizontal by example, provided that their distance D, defined parallel to this dimension of the image, is less than a threshold Dmax, chosen so as to block the space between the characters 10 constituting a word and not to block the space 11 between two brand words.

FIG. 3a is an image obtained after the blocking step line by line, on which the blackened white points between two black points defining the words 6 are marked with gray 12. This gives linear sets of black points, or segments , parallel to the two dimensions of the image, horizontal and vertical, which form a connected component.

This blocking step is followed by a second step of filtering the segments parallel to the first dimension of the image, line after line, over their length L, otherwise called horizontal filtering, respectively vertical, in the case of a horizontal mark. , respectively vertical. The length of the segments selected must be between two thresholds, one minimum Min chosen so as to eliminate the segments too small compared to the length of a word of the mark, and the other maximum Lmaxl chosen so as to eliminate segments of length greater than the total length of the mark.

A third step of filtering on the height of the segments parallel to the second dimension of the image, column after column, otherwise called vertical, respectively horizontal filtering, aims to keep only the vertical segments whose height H is between two minimum Hmin and maximum Hmax thresholds. These two thresholds are chosen as a function of the height of the words of the mark, with the aim of eliminating the words that are too high that do not correspond to the desired marking line, as well as the parasitic blockages between words of the mark. FIG. 3b is an image obtained after filtering on the length and the height of the blocked words 6, in which the blocked areas selected are in gray 12 and the blocked areas not retained due to the height of the words are marked with crosses 13.

FIG. 4 is the image of the connected components 14 represented by greyed words, retained after the steps, line after line, of plugging and filtering over the length of the segments obtained and, column after column, of filtering over their height. This processing of the original image on the horizontal lines and on the vertical columns makes it possible to filter a large part of the segments of this image so as to keep only the components of interest for the rest of the processing. At address 15 in particular, it can be seen in FIG. 4 that the number 45 of the street has been eliminated by filtering along the length, because it is too small. This first phase can very well be carried out in data flow as the image is acquired by specialized electronic cards.

The second phase of the process, which is a phase of localization of the brand in its entirety, includes a first step of labeling the related components, making it possible to obtain a detailed description of each of the components defined by its two dimensions, length and height. corresponding to the dimensions of the rectangle that encompasses it.

A second step of filtering the related components over their length and height aims to retain only the components whose size is compatible with the size of the words of the mark from which the minimum and maximum filtering thresholds are determined, for length and height. This filtering step makes it possible to eliminate a large number of components which correspond either to words of too large size or to pieces of parasitic graphics.

FIG. 5 is the image of the components 14 (grayed words) retained from the image of FIG. 4, after labeling and filtering of the components over their length and their height. At address 15, we see that the definite article "des" associated with "flowers", as well as part of the name of the town, were not retained, being too short or too long.

A third step of merging the related components located at the end of the previous step makes it possible to form lines made up of several words, among which must be the desired marking line. This fusion of the components is carried out according to certain criteria concerning the distance between two consecutive words as well as their difference in alignment, height and orientation. As shown in Figure 6as each component selected in the previous step can be included in a rectangle 18 whose sides are parallel to the two dimensions of the image and whose midpoint is the center of gravity G of the component. orientation of the component is defined by the angle a between the direction 6 of the long side of the rectangle parallel to the first dimension of the image, for example the horizontal, and the direction of the axis of symmetry a of the passing component by the center of gravity G. Figures 6b to 6e represent different relative positions of two components 16 and 17 of an image, these positions being determined from the center of gravity of each component - or midpoint - and its main orientation , depending on the longest side.To merge two consecutive components, the following conditions must be met - the horizontal distance D between the centers of gravity
G16 and G17 is between two minimum thresholds D1 and maximum D2 depending on the length of the words of the mark and their spacing (FIG. 6b); the main orientation difference r between the two components 16 and 17 is less than a threshold rS (FIG. 6c); - the difference in alignment between the two axes of symmetry a16 and a17, that is to say the distance d separating these two axes, is less than a threshold d5 (FIG. 6d); - The difference between the respective heights h16 and h17 of the two components, H = h16-hl7 is less than a threshold h5 (Figure 6e), the thresholds d5 and h5 being determined according to the marking line.

The condition concerning the difference in alignment between the words obliges the marking line to be separated from the information which it makes it possible to locate by a distance greater than the threshold, so that its words are not merged with those of the information. .

Figure 7 is the image on which are marked in gray the lines 19 formed after fusion of the connected components. It can be seen in this image that few lines are formed, in particular at the postal address 15, only two lines have several words after fusion, including the marking line.

Among the lines selected in the previous step, it is necessary to detect the brand whose length and height are known, as well as the number of words that compose it.

This is why the processing continues with a fourth step of filtering the lines on their dimensions, height and length, and on their number of words, step eliminating all the lines not corresponding to a marking line, that is to say - say having a length, a height and a number of words different from those of the mark.

FIG. 8 is the image on which the mark 20 is located, in this case the four words POSTE written under the address and marked with gray.

Finally, after having thus located the mark, it is easy to locate the information sought by analysis of the image area situated in a known manner with respect to the mark. In the case of postal address 15, this is located just above the mark. FIG. 9 is the image on which the information is located, marked in dark gray, from the mark, marked in light gray.

This method of locating information from the addition of a brand has been tested in the field of postal addresses affixed to all kinds of flat physical objects, in particular covered with plastic films representative of the current market. These are the source of numerous parasites appearing on the binary image of objects in the scrolling bench used for sorting mail. Despite this deterioration in the quality of the images and despite their possible complexity, the method according to the invention gives an excellent rate of exact localization of the mark. To prevent the extreme words of the brand from being grouped with parasitic words located on either side of it, when merging words online, it is particularly advantageous to provide white margins on each side of the mark, of sufficient length, that is to say greater than the distance threshold provided for the merger, just as it is necessary to provide a minimum distance between the mark and the information proper to avoid their merger. In the field of postal addressing, in order to ensure proper location of the address block, it is necessary to provide a margin, of at least 5 mm for example, all around the latter. In particular, when the address is printed on a label, this white margin must be observed.

Claims (11)

1. Location mark of information with which it is associated in a complex image comprising words consisting of letters and numbers in block letters and graphics, the relative positions of the mark and of the information being fixed, characterized in that it consists of a line of several words such as - the characters constituting the words of the mark have a font and a size intended to allow blocking of the words; - the space between each word of the mark is greater than the spaces between the words of the image using the same characters as the words of the mark and greater than the blocking threshold; - the size of the words is different from the size of most of the words in the image which are consecutive and likely to form a line on the image; - the number of words is greater than the number of consecutive words likely to form a line on the image and of sizes identical to those of the words of the mark. 2. Registration mark according to claim 1, characterized in that the size of the words is identical for all the words of the mark and the number of words of the mark is at least equal to four. 3. A method of locating information on a two-dimensional image by adding a mark defined according to one of claims 1 or 2, comprising a phase of locating the mark from the binary image of the original image followed by a location of the information from this mark knowing their relative positions, characterized in that the marking of the mark comprises a first phase of filtering the binary image, defined by lines and columns, to keep only the words likely to belong to the brand, consisting of - a first step of blocking the words, line after line, delivering segments of black dots leading to a single related component per word; a second step of filtering the segments along their length, line after line, intended to eliminate the words of length less than the length of a word of the mark and the words of length greater than the total length of the mark; - a third step of filtering the segments on the height, column by column, intended to keep the words of height close to that of the words of the mark; a second phase of localization of the mark on the image consisting of - a first step of labeling the related components obtained in the previous phase; a second step of filtering the related components over their length and their height, intended to keep the words whose size is compatible with that of the words of the mark; - a third step of merging the components located at the end of the first phase, online; and - a fourth step of filtering the lines of merged components on their dimensions, height and length, and on their number of words. 4. Localization method according to claim 3, characterized in that the step of blocking a word (6) of the binary image, consisting of characters made by black dots on a background of white dots, consists of blackening all the white dots (7) lying between two black dots (8) whose distance D, defined parallel to a dimension of the image, is less than a threshold (Dmax), chosen so as to block the constituent characters (10) of a word and not to block the space (11) between two words of the mark. 5. Localization method according to claim 3, characterized in that the second step of filtering the segments, line after line, parallel to the first dimension of the image, takes place over their length (L), which must be between two thresholds, one minimum (Lmin) chosen so as to eliminate the words smaller than those of the mark, and the other maximum (La), chosen so as to eliminate the components of length greater than the total length of the Mark. 6. Localization method according to claim 3, characterized in that the third step of filtering the segments, column after column, parallel to the second dimension of the image, is done on their height (H) which must be between two minimum (Hmin) and maximum (Hmax) thresholds, chosen according to the height of the brand words, in order to eliminate words higher than those of the brand, as well as parasitic blockages between brand words. 7. Localization method according to claim 3, characterized in that, during the second phase of localization of the mark, the first labeling step of the related components defines them by their respective length and height. 8. Localization method according to claim 3, characterized in that, in the second phase, the second step of filtering the connected components is done over their length and their height which must each be between two thresholds, one minimum and the other maximum, defined according to the size of the brand's words. 9. Localization method according to claim 3, characterized in that, in the second phase, the third step of merging the related components appearing consecutively on the image, each defined by its center of gravity and its orientation relative to the horizontal, includes the following conditions - the horizontal distance (D) between the centers of gravity (G16 G17) of two consecutive components (16, 17) is between two minimum (D1) and maximum (D2) thresholds, depending on the length words of the brand and their spacing; - the main orientation difference (r) between the two consecutive components (16, 17) is less than a threshold (fis) ;; - the difference in alignment between the axes of symmetry (a161 a17), in other words the distance (d) separating said axes, is less than a threshold (d5); - the difference between the respective heights (dol6 d17) of the two consecutive components (H = h16-h17 is less than a threshold (h5). 10. Localization method according to claim 3, characterized in that, in the second phase, the fourth step of filtering the lines of merged components consists in comparing the length, the height and the number of words respectively of each line and of the Mark. 11. Method for locating information in the form of a postal address (15) consisting of several horizontal lines placed one above the other, according to one of claims 3 to 10, characterized in that the mark (20) is placed under the last line of the address, the routing line, at a determined distance avoiding any merger and has, at its two ends, a white margin of length greater than the maximum threshold (D2) of distance planned for the merger of related components.