GB2027962A - Automatic mark recognition - Google Patents

Abstract

Information is encoded upon a document (BE) by handwritten marking (M) at one or more positions of a matrix. The matrix is notionally divided into sections each of which should, according to the encoding used, contain a known number of markings. The reading apparatus has a photo- electric scanning unit (OP, SC) which scans the document and supplies video signals (VS) to signal conversion units (SW1, 2, 3) where they are quantised with different threshold values. Scan data (AD1, 2, 3) obtained in this way are further processed in mark recognition units (MSS1 to MSS3) to obtain mark data (MD1, 2, 3). The items of mark data which correspond to differing degrees of density of the mark are subjected to a plausibility check in a plausibility checking unit (PCE), in which the actual number of marks (M) within a mark section, as determined from the sets of mark data (MD) obtained using different thresholds, is in each case compared with the theoretical number in a mark data calculating unit (MIP). That set of mark data (MD1, MD2, MD3) which result in uniformity is output as a plausible read-out result. <IMAGE>

Description

SPECIFICATION Automatic mark recognition The invention relates to the automatic recognition of marks on a machine-analysable document.

The advent of high-speed document readers has had a fundamental influence on data input in automatic data processing systems. They have facilitated the direct processing of data in which the data to be input into the data processing system is directly read out from an original document. In order to achieve the highest possible accuracy of character recognition at favourable costs, mechanical readout facilities have however generally been limited to the recognition of standardised documents.

With this limitation, machine-readable data can be entered onto a data carrier only if so-called coder printers are available at every remote acquisition point. It requires less outlay, is more favourable cost-wise and nevertheless reliable to enter handwritten marks, e.g. by making crosses, in preprinted small boxes on a document printed in machine-readable form. The positions on the document defined by the mark boxes are assigned specific significances, e.g. the digits 0 to 9. The boxes which are to be marked are printed, arranged in rows and columns, in the form of a marking matrix on the document in "drop-out" colours which lie outside of the spectral range of sensitivity of the opto-electric scanning unit and therefore do not influence the mark recognition.Marks entered in handwriting into the mark boxes, such as vertical or horizontal crosses, or also however horizontal lines, are read out centrally by means of an automatic mark recognition device. Devices of this kind can, for many purposes, replace conventional punched cards. Often a device of this kind for the automatic recognition of handwritten marks consists of an attachment to a document sorter.

Then the marking matrix can be read out together with an OCR row in the same document run.

The information content of a mark is obtained, as described above, from the position of the mark on a stipulated blank document. Therefore a device for the automatic recognition of handwritten marks must firstly determine the position of the mark boxes printed on the document relative to the scanning pattern of the device. For this purpose, in mark documents, timing marks can be printed in a row which is arranged parallel to the rows of the marking matrix, each of which timing marks is in alignment with a column of the marking matrix. Our co-pending United Kingdom patent application No. (Serial No. 7926730) (VPA 78P 8031) describes in detail how these timing marks can be used to establish the column and row allocation of scanning data.

As described in the above-mentioned patent application, order to correct the read-out result in the event that mark documents are obliquely inclined relative to the scanning unit, a code column which precedes the marking matrix in the direction of the columns can be analysed. This code column contains code marks which are vertically offset from one another and whose horizontal offset - expressed by the spacing between the columns - is determined and analysed. In the scanning result, all the positions for the mark boxes of a column of the marking matrix are corrected electronically. This solves the problem of a clear assignment of scanning data to a specific mark box or, quite generally, to the scanning positions of the marking matrix. The more accurately this can be done, the less danger is there that marks will be assigned to an incorrect position or to a plurality of positions.

However, marks, in particular crosses, can vary considerably in contrast, size, shape and position within the mark box. In mark readers this variation breadth must be taken into account since the marks are entered in handwriting and will consequently express the peculiarities of the writer and of the writing implement he is using. In addition it must be taken into account that the documents become dirty during use.

In automatic character recognition it is normal to distinguish useful signals from interference signals by deriving from the outer field of the scanning point a signal for controlling the quantisation threshold. This means a contrast-dependent conversion of the analogue video-signals into digital scan signals. In mark readers it is thus possible to control to some extent one of the factors which influence the variation breadth, namely changing contrast.

Nevertheless it is still extremely difficult satisfactorily to recognise-the marks which in themselves are extremely simple, as the other influential factors can hardly be controlled by digitalisation of the video signals with a variable threshold value.

According to the present invention there is provided an apparatus for reading a document upon which information has been encoded by the making of markings at ones of a plurality of predetermined positions of a marking matrix, comprising scanning means arranged in operation to scan the document and produce analogue image data in respect of elements defined by the rows and columns of a scanning matrix, a plurality of signal conversion units each arranged in operation to form from said analogue image data digitalised image data which assumes a first or second value according to whether the analogue image data is above or below a predetermined threshold value, the signal conversion units operating with mutually different threshold values to produce different sets of digitalised image data, mark analysis means arranged in operation for each set of digitalised image data to produce, in accordance with predetermined criteria, mark data indicative of whether, in respect of each of said plurality of predetermined positions, that set of image data represents the presence or absence of a marking, plausibility checking means arranged in operation for at least some of said sets of mark data in response to information supplied thereto indicating, for the type of document being read, notional division of the marking matrix into sections and the theoretical number of markings which should appear in each such section, to compute for each such section the number of markings indicated by the relevant set of mark data as being present therein and to compare those numbers with the corresponding theoretical numbers and output selector means responsive to the results of said comparison to select and output that set of mark data which appears plausible.

In another aspect the invention provides a method for the automatically recognising handwritten marks on a machine-analysable document, wherein mark boxes printed in a "drop-out" colour are optically scanned columnwise, video signals are formed in an opto-electrical transducer from which are derived digitalised items of scanning data which are to be further processed in serial fashion and which can be clearly locally assigned to the mark matrix, from which a mark statement is obtained by combining a plurality of scanning elements, in which the serially occurring video signals are fed to a plurality of parallel signal conversion units where they are quantised with stepped threshold values to form items of scanning data and these are further-processed in parallel and prepared to form items of image data, that in a scanning pattern, in which the mark boxes extend horizontally over a plurality of scanning colums and vertically over a plurality of scanning rows, the image data for each mark position is added row-wise in the column zone which is covered by the mark box so that with increasing distance from the centre of the mark box the individual image data elements are evaluated in stepped fashion with lower weighting that a plurality of row sums which follow one another in the column direction are in each case multiply combined to form image element combinations in such manner that zones of vertically adjacent mark boxes are also covered and a statement concerning a mark is derived from these determined image element combinations, that these items of mark data which have been formed in this way from the image data are intermediately stored in a buffer store and, in parallel fashion by adding the marks, established in each mark section, within the same quantisation stage and carrying out a comparison with a theoretical value, stored in a register, for the number of marks per mark section it is determined which of the read-out results of the various quantisation stages appears plausible, and that this read-out result from the intermediately stored mark data and output by way of result.

Thus a plurality of quantisation thresholds are employed which are graded and with which analogue, serial video signals may be digitalised. The various items of scanning data which are obtained in this way and which each produce a black-white pattern are processed and analysed in parallel. In order to obtain mark data from the scanning data prepared to form image data, the scanning pattern is selected to be such that a plurality of scanning rows and columns pass over a mark position. This means a high local resolution and, considered in the row direction, allows the items of scanning data for a mark position to be evaluated differently in dependence upon their position in edge zones or central zones.

Influences of excessively large marks in horizontally adjacent mark boxes can be reduced in this way.

Similar influences of marks from vertically adjacent mark boxes can be taken into account in that the scanning data of the boxes which are vertically adjacent to the mark box currently being considered are included in the mark decision. In this way it can be established for example whether a density in the mark box currently being analysed emanates from an excessively large mark in a neighbouring box. The mark statement is based on several row configurations, i.e. image element combinations in various rows which possess a different relative position to the mark track. The image element combinations which are to be evaluated as mark are stored in a read-only memory which supplies the yes/no statement.

As the scan data of each quantisation stage are processed and analysed separately, several graded mark statements are produced for each mark box. As only one single yes/no result is to be output, a result selection must be made using a plausibility criterion. A plausibility criterion of this kind is available whenever the mark document can be notionally divided into mark sections for which a specific theoretical number of marks is prescribed. However, plausibility criteria change with the nature of the document. If a number of different types of document are to be processed in a mark reader, these criteria must be changed for each individual document. A code column which precedes the marking matrix in the scanning direction can be used to effect this automatically.The nature of the document may be defined by means of combinations of printed code marks in this column, by which means it is possible to derive the criteria concerning the size and position of the mark sections and the theoretical number of marks which occur therein.

These criteria then determine the plausibility control programme in which the number of marks in a mark section actually read out during the various quantisation stages is compared with the theoretical number.

The items of mark data of those quantisation stages in respect of which identity between theoretical and actual value has been determined, are considered as the plausible read-out result and output for further processing. If mark documents are involved in respect of which this plausibility control cannot be carried out, the read-out result of a central quantisation stage is output.

The criterion for plausibility of a set of mark data may, for example, be that the said comparison indicates equality between the theoretical and actual number of marks in each section - or that this is the case within an allowable margin of error. Where the computation and comparison is carried out only on mark data obtained using high and low thresholds, it may be arranged that, in the event that these results are not plausible, data obtained using an intermediate threshold (upon which a plausibility check has not been carried out) is then selected for output.

An exemplary embodiment of the invention will be explained in detail in the following making reference to the drawings in which: Figure lisa block circuit diagram of a device designed in accordance with the invention for the automatic recognition of handwritten marks; Figure 2 illustrates one of the mark recognition units of the device of Figure 1; Figure 3 illustrates a buffer store for the intermediate storage of the mark data which associated input circuits and output circuits; and Figure 4 is a block diagram of the plausibility checking unit of the device of Figure 1.

Figure 1 is a block circuit diagram of a device for the automatic recognition of handwritten marks on a machine-analysable document BE. This document contains a number of marks M, in the form of crosses, in a marking matrix constructed from rows and columns. At the inter-section points between the rows and the columns the marking matrix is provided with mark boxes (not shown) for marks M to be entered by hand.

The mark boxes are printed in drop-out colours which lie outside the spectral range of sensitivity of the scanning unit and which have not, therefore, been shown. Above the marking matrix is arranged a row of timing marks TM which are printed in alignment with the columns of the marking matrix. Thus these define the horizontal position of each mark column within the marking matrix. The vertical coordinates of the individual mark rows can also be derived from the vertical position of the timing marks on the machine-analysable document. Thus when the row and column zones are known individual sections can be specified within the marking matrix.

The marking matrix is read out in a direction opposite to the direction of conveyance, indicated by an arrow A, of the document and the first column to be read out is always a column to be read out is always a column provided with code marks CM. These code marks consist of a series of line marks in specific row positions and define the nature of the document. Here it should merely be noted that additional code marks CM can also be used to correct the position of the mark boxes in the case of an oblique document.

A document BE of this kind is to be machine analysed in a readout station in the input section of a document sorter. The document BE is conducted past a scanning window where it is illuminated by a light source (not shown). Light reflected by the document is fed via scanning optics OP to a opto-electric transducer SC comprising a row of photo-diodes arranged parallel to the column direction of the marking matrix.The photo-diodes are cyclically interrogated in serial fashion by a scanning circuit in the opto-electronic transducer SC and the resultant video signal VS is fed to signal conversion units SW1 to SW3. In order to eliminate the influence of the strength of illumination and the differing reflective properties of the document surface, the brightness of the character background, i.e. the current maximum white value in the row of the image point which is to be scanned, is taken into consideration in the quantisation of the video signals VS to form an image pattern composed of black-white values.From the brightness of the character background which is established, three black-threshold values, staggered in level, are derived with the aid of which three different items of scan data AD1 to AD3 are obtained from the original video signal VS.

Afourth, high threshold is used to evaluate the timing marks TM and code marks CM which are printed so as to be strong in conrast, and the items of scan data AD4 are produced in a further signal conversion unit SW4.

The positional information for the mark boxes is derived from the scan data AD4 relative to the timing marks TM in a pre-processng unit PREP and is fed to a parameter store PARM. The scan data AD4 relating to the code marks CM are used to measure any oblique positioning of the document BE and accordingly an electronic correction in respect of the position information for the mark boxes is effected. Further details are given in our U.K. patent application No. referred to above.

The items of scan data AD1 to AD3 which are pre-processed in dependence upon the items of scan data AD4 are temporarily stored, as items of image data BD1 to BD3, in an image pattern store DSM. Items of mark data MD1, MD2, MD3 are then formed from the items of image data. This is carried out separately for the image data BD1, BD2 and BD3 of the three quantisation stages in one of three mark recognition units MSS1, MSS2 and MSS3.

Many types of documents marked in handwriting, e.g. betting slips, allow a plausibility check of the read-out results since a specific number of marks must always occur in specific sections of the marking matrix. The combination of the printed code marks CM indicates the nature of the document so that, with the aid of criteria derived therefrom, it is possible to check which of the three sets of results appears plausible.

This checking of the mark data is carried out in a plausibility checking unit PCE. In dependence thereupon, the items of mark data MD1, MD2 or MD3 which appear plausible are selected and transferred to a mark output unit MOP.

In order to provide for the intermediate storage of the mark data during this check, the plausibility checking unit PCE contains a buffer store FIFO which operates in accordance with the first-in-first-out principle. The plausibility check itself is carried out by a mark data calculating unit MIP in which the number of established marks M within a section of the marking matrix is determined from the mark data MD1 to MD3 and compared with a theoretical value determined by the nature of the document. With documents of different types, the relevant theoretical values and further control criteria, in particular control signals for section sizes, must first be determined. This is effected by a programmed control unit PCU which analyses the mark data MD3 of the code column supplied by the mark recognition unit MSS3. An assigned control programme in the PCU is selected in accordance with the code combination discovered. The PCU now controls the calculating unit MIP whose results in turn control the output multiplexer OMUX.

Figure 1 also shows internal checking devices which serve to check the functioning reliability of the devices described. Following each switch-on of the device and following a fault normalisation, a programme for recognising a simulated document with a full marking matrix is firstly run. All the data required for this purpose is taken from a read-only memory - the simulation store S-ROM. These items of data are also analysed as explained above. Mark results are then compared with stored theoretical values in the simulation store S-ROM. In the absence of conformity a fault signal is emitted and the recognition device is blocked from normal operation.

This general survey will now be followed by a more detailed discussion of how items of mark data MD1 to MD3 are derived from the items of scan data AD1, AD2 and AD3. The scanning pattern is much finer than the pattern of the mark matrix so that one mark position gives rise to 6 x 12 items of scan data and, following information reduction, 3 x 12 items of image data. Thus following information reduction one row of the marking matrix still comprises three image rows of the standard document, whereas the horizontal resolution is much greater as every mark column is scanned in a plurality of horizontal steps and here no data reduction is carried out. Lines, vertical croses and, in particular, horizontal crosses are to be recognised by way of marks.

Figure 2 illustrates in detail one of the three mark recognition units MSS. Also shown here is the image pattern store DSM which stores the items of image data BD1 to BD3. The items of image data BD1 are transferred in serial fashion into the mark recognition unit MSS1 where the items of image data of an image row are added over a width often scanning columns centred on a mark position. This means that the scanning columns to be assigned to the left and right edge of a mark position are now suppressed. A document is scanned with 128 vertically adjacent photo-diodes. Thus, following a row reduction of 2 to 1 in the preprocessing unit PREP 64 items of image data are still assigned to each scanning column and are fed in serial fashion column-wise to the mark recognition unit MSS1.In order to be able to add the serially supplied items of image data BD1 of 10 scanning columns in the row direection in an adder circuit ADD, an accumulator ACC is required which consists of four shift registers each having 64 positions.

The row sum formation in the adder circuit ADD is governed by the following special conditions, apart from the gating out of the edge columns by a release signal MCO: the items of image data of the individual scaning columns are differently evaluated in accordance with their position relative to a mark point. The items of image data in the six central scanning columns of the analysis zone are included in the row sum with a value "1" whereas the remaining four edge columns are given a correspondingly lesser weighting of "0.5".

For this purpose the adder circuit ADD is additionally supplied with a column-related value signal WCS. The sum value is expressed as a four-bit binary value which has a maximum of 7.5. Thus each image row of the analysis zone is assigned a binary value of this kind which comprises four bit positions and which is fed to the adder circuit ADD from the accumulator ACC and to which the current value from the flow of image data BD1 is added.

Here it should also be noted that black image data having a digital value of "1" or "0.5" is added whereas white image data having the same absolute binary value is subtracted. However, the subtraction takes place in such a way that the row sum value cannot be negative and consequently it is unnecessary to distinguish between signs. In practice this means that the row sum calculation is influenced by white gaps in the row of the analysis zone.

The adder circuit ADD is also connected to a comparator circuit MCOMP which serves to compare the current row sum value with a maximum value stored in a maximum value store MAX. If the comparison between the two binary values indicates that the current row sum value is higher than the previous maximum, the comparator circuit MCOMP activates a conjunctive logic circuit UG1 and the current row sum value is input into the maximum value store MAX. Like the accumulatorACC, this maximum value store MAX likewise consists of four shift registers each of 64 positions, and therefore can accommodate the maximum sum values of all 64 image rows for an overall mark column.

Three consecutive maximum row sum values are in each case intermediately stored in three row sum registers SREG1, SREG2 and SREG3 which are connected in cascade to the maximum value store MAX. By definition three image rows belong to one mark row or track. Therefore these row sum registers contain, in the form of the maximum values, all the information concerning the image data BD1 of an analysis zone required to be able to recognise any cross mark M. All the possible and permissible combinations of these row sum values are stored in a mark recognition store MCM. The contents of the row sum registers serve as the addresses of said store. With regard to a mark box, three analysis positions are provided. In a central position the image rows to which the currently stored row sum values relate, precisely cover the mark track.

In an upper position two image rows lie above the track and similariy in a lower position two rows lie below the track. The two image rows from a neighbouring track which are taken into account permit one to distinguish whether a density in the upper or lower image row of the track which is currently to be analysed originates from a very small, badly positioned mark M or from an excessively large mark in the neighbouring box.

For each of the 64 image row positions, the mark recognition store MCM supplies, in parallel, three mark statements which are derived from the three aforementioned analysis positions. However, only one of the three mark statements applies in each image row position. For this purpose these three digital statements "mark present" or "no mark present" are fed in parallel to a mark multiplexer MMUX. The latter is fed by the parameter store PARM with a two-bit selector signal VTM which is derived from the vertical position of the timing mark TM of this mark column. In this way each processing step is assigned one of the three possible positions relative to the tracks. Thus three mark messages which operate disjunctively are emitted consecutively from the output of the mark multiplexer MMUX in respect of a mark box.

Figure 2 illustrates only one mark recognition unit MSS1, although, as mentioned above, the recognition device contains three such mark recognition units MSS1, MSS2 and MSS3 each of which process the items of image data BD1 to BD3 in parallel to form mark data MD1, MD2 and MD3.

Figures 3 and 4 illustrate parts of the plausibility checking unit PCE whose function is to select a quantisation threshold for the information output from the items of mark data MD1, MD2 and MD3. A plausibility check can take place only if it can be stated, in respect of a specific type of document, how many marks are to be contained on such a document or in a part of its mark matrix.

The nature of the document is coded in the combination of the code marks CM. This code also contains the information regarding the division of the mark matrix into a plurality of sections of equal size for each of which a plausibility check is to be effected. An application of this kind consists for example of betting slips which are to be automatically analysed.

On the basis of the code contained in the code marks CM the plausibility checking unit PCE must start a specific checking plan which is dependent upon the data for the section size with a known number of mark rows and mark columns and a theoretical number of crosses M marked within such a section.

However, the plausibility check is not carried out for the mark data MD1 to MD3 of all three quantisation stages. It is in fact sufficient to check the mark data MD1 of the lowest quantisation stage and the mark data MD3 of the highest quantisation stage. The plausible mark data is selected on this basis. If, however, neither of the two read-out results appears plausible, the mark data MD2 of the central quantisation stage is output.

This also applies in the case of document types for which a plausibility check cannot be effected. In these cases it must be assumed that the items of mark data MD2 derived from the image data BD2 of the central quantisation stage represent the read-out result involving the least fault frequency.

Figure 3 shows the mark recognition units MSS1 - 3 and the buffer store FIFO which serves for intermediate storage of the mark data MD1, MD2 and MD3 during the plausibility check. This buffer store consists of three sub-stores FIFO1, FIFO2 and FIFO3. The designations of the sub-stores indicate their mode of operation. In order to fully exploit the capacity of commercially available storage modules, the items of mark data MD1, MD2 and MD3 which are offered in serial fashion from the three mark recognition units MSS1, MSS2 and MSS3 are combined, two bits at a time, via two OR gates OG1 and input to one of two shift registers SPW1 and SPW2 which serve as serial-to-parallel converters.These each store six bits and they serve as input registers for the three sub-stores FIFOl, FIFO2 and FIFO3. The latter are constructed in such manner that one stored word comprises four bit positions. Therefore the output lines of the serial-to-parallel converters SPW1 and SPW2 are each divided between two of the sub-stores.

The transfer of the stored words into the sub-stores FIFOl, FIFO2 and FIFO3 is synchronised by a write pulse SF. This is produced by an information element counter ISC which is synchronised at the beginning of each checking process and each time an information element is transferred receives a transfer pulse IS as counting pulse. Following every six information elements it emits an output pulse which is combined with a timing pulse TP in a second AND gate UG2 to form the write pulse SF.

The recoded mark data are converted back at the output of the three sub-stores FIFOl, FIFO2 and FIFO3.

The output register consists of two parallel-to-serial converters PSW1 and PSW2. The serial outputs of these two parallel-to-serial converters are each connected to an input of a further OR gate OG2 and to an input of a further AND gate UG3. Then the serial, but now delayed, items of mark data MDA1, MDA2 and MDA3 are again available at the three outputs of this decoding network.

The plausibility check is carried out whilst the items of mark data MD1, MD2 and MD3 pass through the FIFO store. Figure 4 shows in detail the mark data calculating unit MIP required for this purpose, the programmed control unit PCU and the output multiplexer OMUX. The programmed control unit comprises a code word register CWR, a start address store SAM, a control address counter PCC, a control store PCM, a control register CSR, a column counter CCT, and a theoretical number register MSR. The mark data calculating unit MIP comprises two mark data adders MAD1 and MAD3, buffer registers BR1 and BR3, accumulator stores MAC1 and MAC3, comparator circuits PCOMP1 and PCOMP3 and result registers PRL1 and PRL3.

Since these parts of the plausibility checking unit PCE cannot be fully described in terms of function without describing specific control procedures which take place within a time pattern, this block circuit diagram in Figure 4 also shows a plurality of control signals which characterise the cooperation between the individual sub-circuits.

For the plausibility check, the programme control unit PCU must receive an item of information regarding the type of document in question. During the preprocessing items of code data CD are obtained from this information contained in the code marks CM and are transferred in serial fashion to the code word register CWR which is designed as a shift register. The satisfactory transfer of the items of code data CD is subject to two additional conditions: the code word register CWR must be reset by a normalising signal NO which occurs at the beginning of each scanning procedure of a document BE. The normalising signal NO is also employed elsewhere in the plausibility checking unit. In addition, items of code data CD must be distinguished from items of mark data MD: this is possible by reference to the scanning columns in which they occur.In the present case an accompanying signal CS defines a transfer cycle for the data of the code column. The length of this cycle is determined by the scan duration of a marking column. The items of data to be assigned to a mark position then occur at regular intervals within such a transfer cycle. The time interval is defined by an information pulse train lNF. AND-ed together, the two last-mentioned signals form the timing signal for the code word register CWR.

Following the transfer cycle for the code data CD, the code word register CWR contains all the information required regarding the nature of the document in question. The code word register forms the address register for the start address store SAM in which the contents of the code word register CWR is decoded into four address signals SAMO to SAM3. The two higher-value address signals SAM2, SAM3 contain a sub-address for the control store PCM and the main address for the control store PCM is indirectly derived from the two other address signals.

A knowledge of the construction and function of the control store PCM is necessary in order to be able to explain this process. The control store is a read-only memory which, in dependence upon the nature of the document, serves to make available, in the correct time sequence, the various control signals for the execution of the plausibility check. For this purpose, for every type of document, it comprises a section containing control words, the start address of this section being selected at the beginning of a transfer cycle for mark data of a column of the mark matrix. In each of these sections the first store word defines the number of mark columns which are assigned to a mark section. Atheoretical number, i.e. the permissible number of marks M per section, is stored at the second address.

Finally the third stored word contains the first of the actual control words. Here it should firstly be mentioned that different mark sections on various documents can comprise different number of rows of the mark matrix. With column-wise scanning of a document BE and associated column-wise processing of mark data MD1 ... MD3, consecutive mark positions within a column can thus belong to different mark sections.

Expressed in other words, mark sections which are assigned to mark columns and are arranged one above another on the document form a section group whose associated mark data is fed to the plausibility checking unit PCU in serial fashion where, however, it is processed in parallel fashion. The linking of individual items of mark data MD1 ... MD3 to form a specific mark section within a section group is determined by these control words in the control store PCM.

As mentioned above, the higher-value address signals SAM2, SAM3 form only a part of the address of the control store PCM. The control address counter PCC is provided to complete the store address. At the beginning of a transfer cycle of mark data MD1 ... MD3 of a mark column, the corresponding start address must be input into this counter. The start address is derived from the second address signal SAM1. This signal is transferred in that a loading input LD is activated by a release signal which is produced as a result of an AND-link of the first address signal SAMO and an accompanying signal CYO which occurs at the beginning of a transfer cycle.This release signal inhibits the counting process of the counter and enables the transfer of the data signals which are offered at the data inputs and which, following the next timing pulse TP at the counting input C, are available at the output. This establishes the start address for the control store PCM.

In each transfer cycle the control address counter PCC must firstly address the two above mentioned addresses of the control store PCM with the items of mark section data, which are dependent upon the nature of the document, and must then address the control addresses. Therefore its enable input EN is supplied with four control signals via an AND gate having an inverted output. These control signals consist of the inverted information pulse train I N Fr the third output signal CSR2 of the control register CSR, the inverted output signal CTO of the column counter CCT, and an inverted signal RCY for an erasing cycle. The function of the two last control signals will become clear in the following description of the control register CSR and of the the column counter CCT.

In order to understand the function of the control register CSR it must be borne in mind that the output signals PCMO to PCM3 of the control store PCM are supplied to the column counter CCT, the theoretical number register MSR, and the two accumulator stores MAC1, MAC3. The control register CSR now assumes the function of switching through these output signals under pulse control to the first two units and of preparing the control address counter PCC for this purpose.

The control register CSR is a 3-bit shift register which is again reset by the normalising signal NO at the beginning of the scanning of a document BE. It receives one bit at the beginning of a transfer cycle which is characterised by the accompanying signal CYO. This control bit is advanced with every timing pulse TP until the output signal CSR2 at the third output of the control register CSR is positive. This output signal is combined with the signal CYO for the beginning of a transfer cycle and returned to an inhabit input INH of the control register CSR so that the latter does not change its condition until the beginning of the next transfer cycle.

At the three pulse times which are each governed by a timing pulse TP, the control register CSR thus emits the output signals CSRO, CSR1 and CSR2. The first output signal CSRO controls the transfer of the first control word from the control store PCM into the column counter CCT. The second output signal CSR1 controls transfer of the second control word out of the control store PCM. The third output signal CSR2 allows the control address counter PCC to count from the start address until the third control word in the control store PCM independently of the transfer of an information element with the information pulse train INF. During the remainder of a transfer cycle the contents of the control register CSR remains unchanged.

The column counter CCT represents a counting circuit which is constructed similarly to the control address counter PCC. Its function is to count the number of columns which belong to a mark section or to a section group and have already been transferred to the plausibility checking unit PCE. As soon as the maximum number of columns for a mark section or section group dependent upon the nature of the document has been reached, the column counter is to emit an output signal CTO characterising the termination of the transfer cycles relating to a mark section or section group. As this number of columns is dependent upon the nature of the document, and this output signal CTO is not emitted until the counter has reached a maximum count, at the beginning of the analysis of a mark section it is set at a specific starting state.This starting state is contained in the first control word of a selected address section in the control store PCM. In order that this data word may be transferred, the loading input LD is activated by the first output signal CSRO of the control register CSR. However, this does not apply to an erasing cycle, which is accompanied by the signal RCY. This erasing cycle is of special significance to the mark data calculating unit MIP in relation to which it will be explained. Following the beginning of each transfer cycle for mark data MD1 ... MD3, the column counter CCT is released by a signal CYO' which is delayed in relation to the corresponding signal CYO, and counts onwards by one bit position with the next timing pulse TP. As soon as it has reached a maximum count, it emits the output signal CTO.This output signal not only controls the release of the control address counter PCC, but is also linked with the normalising signal NO which occurs at the beginning of the scanning of a document BE to form the aforementioned signal RCY for an erasing cycle.

The theoretical number register MSR in the programmed control unit PCU is likewise connected to the data outputs of the control store PCM and receives the latter's output signals PCM2 when this is released by the second output signal CSR1 from the control register CSR. The analysis of the mark data MD1, MD3 with regard to the plausibility check is now carried out in the mark data calculating unit MIP. Here, as mentioned, a plausibility check is carried out only in respect of the mark data MD1 and MD3 of the first and third quantisation stages. As may be gathered from Figure 4, consequently the corresponding assemblies of the mark data calculating unit MIP have been provided in duplicate. Therefore the description will be restricted to an explanation of the processing of the mark data MD1 of the first quantisation stage.

In the mark data calculating unit MIP these items of mark data MD1 supplied in serial form are to be added separately for each mark section and, on the conclusion of the transfer of a mark section or section group, are to be compared with the theoretical number contained in the theoretical number register MSR, whereupon the result is output into the result register PRL1. For this purpose the serially presented items of mark data MD1 are fed to the assigned mark data adder MAD1. The latter is simultaneously supplied from a buffer register BR1 with the current sub-total of marks so far determined in respect of this mark section. A value which occurs on the transfer of a mark point is added to this previous value and the new value is transferred into the buffer register BR1.

In the control section the transfer of a mark point is pulsed by the information pulse tran INF with which the control address counter PCC counts onwards by the value "1". In the control store PCM this addresses the next control word which contains an address for the accumulator store MAC1. If the next mark row in the current column again relates to the mark section which was previously in question, this new control word of the control store PCM also contains the previously selected address of the accumulator store MAC1. If, however, the following mark row belongs to the next mark section of the section group which is being processed, the control word read out contains the next address for the accumulator store MAC1. Thus the mark data added MAD1 is supplied with the previously established number of marks for this new mark section which can be added to the current mark.

This process is repeated during a transfer cycle for a mark column. It is possible that the marks encountered may fall into more than one mark section. Column by column this process is continued until the column counter CCT has reached its given value and thus the last mark column belonging to the mark section or section group has been analysed. It will now be clear why the column counter CCT is released by the signal CYO' which is delayed relative to the accompanying signal CYO. This ensures that the information of the last mark column forming part of the mark section is also correctly added.

Although not illustrated in Figure 4, the output signal CTO from the column counter CCT is held for the period of one cycle. During this cycle the theoretical-actual comparison is carried out, during which the control address counter PCC continuously counts upwards so that all the addresses of the accumulator store MAC1 are addressed. The items of data emitted from the output of this store are fed to the comparator circuit PCOMPI and compared with the corresponding theoretical value from the theoretical number register MSR.

The comparison result is transferred into the result register PLR1 which is constructed as a shift register.

This comparison cycle is followed by an erasing cycle for the accumulator store MAC1. The erasing cycle is characterised by the signal RCY. Again in this cycle all the addresses of the accumulator store MAC1 are addressed. Now, however, the contents of the buffer register BR1 are maintained at "zero" by the accompanying signal RCY present at its resetting input R. Thus a "zero" is entered consecutively into all the store positions of the accumulator store MAC1. Following the erasing cycle the mark data calculating unit MIP is ready for a new section check.

The result relating to the check section or section group is known. Thus the items of marked data MD1, MD2 or MD3 as the case may to which appear plausible can now be output. During this output cycle the delayed items of mark data MDA1, MDA2 and MDA3 are fed from the sub-stores FIFO1 to FIFO3 to the output multiplexer OMUX which is controlled appropriately for each mark section by the control signals supplied from the two result registers PLR1 and PLR3. In respect of each mark section, all those items of mark data of the quantisation stage are output for which a plausible read-out result has been established. Documents whose mark contents do not permit a plausibility check to be effected are characterised by a corresponding combination of code marks CM in the same way as other document types. The output cycle is then controlled in such manner that the items of mark data MDA2 of the second quantisation stage are output.

The apparatus described thus provides means whereby, even in the case of handwritten marks; local variations in contrast, size and shape do not substantially impair the read-out accuracy and permit automatic document processing with a low fault rate.

For ease of reference, a list of the symbols used in the description is given below: List ofReferences Figure 1 BE document M marking TM timing marks A direction of conveyance of document CM code marks OP scanning optics SC opto-electric transducer VS serial video signal SW1 SW4 signal conversion units PREP pre-processing unit PARM parameter store DSM image pattern store MSS1 mark recognition unit for image pattern/stage 1 MSS2 mark recognition unit for image pattern/stage 2 MSS3 mark recognition unit for image pattern/stage 3 PCE PLAUSIBILITY CHECKING UNIT MOP mark output unit FIFO buffer store OMUX output multiplexer MIP mark data calculating unit PCU programmed control unit Figure 2 ADD adder circuit ACC accumulator MCO release signal for mark column WCS value signal MCOMP comparator circuit MAX maximum value store UG1 AND gate SREG 1,2,3 row sum registers MCM mark recognition store MMUX mark multiplexer VTM selector signal (vertical position of the timing marks) MD mark data Figure 3 FIFO1,2,3 sub-stores for mark data OG1 first or gates SPW1 ,SPW2 shift registers/serial-to-parallel converters SF write-in pulse ISC information position counter IS transfer pulse TP timing pulse UG2 second AND gate SF write-in pulse - PSW1 ,PSW2 parallel-to-serial converter OG2 further OR gate UG3 furtherANDgate MDA1,MDA2,MDA3 serial output data of the FIFO store Figure 4 CWR code word register SAM start address store PCC control address counter PCM control store CSR control register CCT column counter MSR theoretical number register MAD1,MAD3 mark data adder BR1,BR3 buffer register MAC1 mMAC3 accumulator store PCOMP1,PCOMP3 result register CD code data NO normalising signal CS - signal accompanying code data transfer cycle INF information clock pulses SAM0 SAM3 address signals CYO signal accompanying beginning of a transfer cycle of mark data TP timing pulse PCMO...PCM3 output signals of control store CSR0 CSR2 output signals of control register CTO output signal of column counter RCY accompanying signal for erasing cycle CYO' delayed accompanying signal for transfer cycle.

Claims (18)

1. An apparatus for reading a document upon which information has been encoded by the making of markings at ones of a plurality of predetermined positions of a marking matrix, comprising scanning means arranged in operation to scan the document and produce analogue image data in respect of elements defined by the rows and columns of a scanning matrix, a plurality of signal conversion units each arranged in operation to form from said analogue image data digitalised image data which assumes a first or second value according to whether the analogue image data is above or below a predetermined threshold value, the signal conversion units operating with mutually different threshold values to produce different sets of digitalised image data, mark analysis means arranged in operation for each set of digitalised image data to produce, in accordance with predetermined criteria, mark data indicative of whether, in respect of each of said plurality of predetermined positions, that set of image data represents the presence or absence of a marking, plausibility checking means arranged in operation for at least some of said sets of mark data in response to information supplied thereto indicating, for the type of document being read, notional division of the marking matrix into sections and the theoretical number of markings which should appear in each such section, to compute for each such section the number of markings indicated by the relevant set of mark data as being present therein and to compare those numbers with the corresponding theoretical numbers and output selector means responsive to the resu Its of said comparison to select and output that set of mark data which appears plausible.

2. An apparatus according to claim 1 including a buffer store for temporary storage of the mark data.

3. An apparatus according to claim 1 or 2, in which the mark analysis means comprises preprocessing means arranged in operation to ascertain the relationship between the scanning matrix and marking matrix, adder means arranged, for each successive marking column to add the digitalised image data corresponding to locations of each scanning row within that marking column with relatively lower weighting for the locations distant from the centre of the column to form a row sum, and mark recognition means arranged in operation to analyse each row sum in combination with the row sum of adjacent rows in the same column to produce said mark data.

4. An apparatus according to claim 3, in which the adder means is so arranged that according to whether the digitalised image data in respect of an element indicates the presence or absence of marking, a value according to said weighting is respectively added or subtracted, provided that the sum would not thereby become negative.

5. An apparatus according to claim 3 or 4, in which the adder means comprises an adder having a first input arranged to receive digitalised image data, the image data for each scanning column being supplied thereto scanning row by scanning row, a second input arranged to receive one at a time from an accumulator intermediate row sums and an output connected to the accumulator whereby the adder adds the digitalised image data in respect of each location in turn of a scanning column to the current intermediate row sum corresponding to the relevant scanning row and replaces the intermediate row sum with an updated intermediate row sum, a maximum value store and comparator means arranged to compare each updated intermediate row sum with the current maximum value stored in the maximum value store relating to that row and to transfer the updated row sum into the maximum value store whenever the updated row sum exceeds the previous maximum value.

6. An apparatus according to claim 5, in which the adder is a programmable read-only memory.

7. An apparatus according to claim 5 or 6, in which the mark recognition means comprises a plurality of row sum registers to which upon completion of the row addition process in respect of a marking column said maximum row sum values are fed in sequence such that the registers contain maximum row sum values in respect of a plurality of adjacent rows, and a mark recognition device arranged in operation to recognise sequences of row sum values passing through the row sum registers as representing the presence or absence of markings in respect of the marking positions, whereby to produce said mark data.

8. An apparatus according to any one of the preceding claims, in which, for reading of a plurality of different types of document each of which can possess mark sections of differing format and a varying theoretical number of marks, and includes a code column which precedes the marking matrix in the row direction and contains code marks indicating the type of the document, the apparatus including means operable to produce digitalised image data from the analogue image data in respect of the code marks, to derive therefrom said information indicating the division of the document into sections and the theoretical number of markings which should appear in each section, and to supply said information to the plausibility checking means for controlling the sequence of the checking of items of mark data in respect of a plausible read-out result specifically for the type of document in question.

9. An apparatus according to any one of the preceding claims, in which the plausibility checking means comprises, for each set of mark data in respect of which said computation and comparison are to be carried out, a mark data adder arranged in operation to serially receive the relevant mark data to add each item thereof to a subtotal stored in an accumulator store in respect of the relevant section and to store the updated subtotal in the accumulator store in place of the previous subtotal, a theoretical number register arranged to receive said information as to the theoretical number of markings for each section a comparator arranged in operation to compare, in respect of each section, the final total formed by the mark data adder with contents of the theoretical number register and to store the comparison result in a result register whereby the result register contains the comparison results in respect of all the sections.

10. An apparatus according to any one of the preceding claims comprising three said signal conversion units operative with a lower, an intermediate and a higher predetermined threshold value whereby, in use, respectively first, second and third sets of mark data are derived by the mark analysis means, wherein the plausibility checking means comprises a respective mark data adder, comparator and result register only in respect of the first and third sets of mark data and wherein the output selector means comprises an output multiplexer responsive to the contents of the result registers to select for output whichever of the first and third sets of mark data is indicated as plausible, or, if neither is so indicated, the second set or mark data.

11. An apparatus according to claim 10, including a programmed control unit assigned to the plausibility checking means and comprising a code word register into which, prior to the analysis of the mark data of a document, the established code data is input in serial fashion where it is held in static fashion until the next document is analysed, a start address store which is connected to the code word register, consists of a programmable read-only memory and serves to recode the code data into a start address for a control store which likewise consists of a programmable read-only memory and which, for each type of document, possesses a store section storing a series of control words, and which is assigned, as address register, a control address counter which is to be set at a start address by the start address store, and whose read-out outputs, which carry output signals representing said information, are connected to supply a pre-adjustable column counter, the theoretical number register, and address inputs of the accumulator stores.

12. An apparatus according to claim 11, in which in each section of the control store which can be selected by a store address, in the first control words the control store contains codes relating to the number of mark columns assigned to a mark section of the relevant document type, and the associated theoretical number, and under the following addresses contains control words for addressing the accumulator stores.

13. An apparatus according to claim 12, in which the control address counter is so arranged that at the beginning of a transfer cycle for mark data it can be reset to the start address by a reset signal, whereupon it recommences to count upwards in stepped fashion under pulse control until the first control word with an address for the accumulator store has been addressed, in which the control unit further comprises a control register which serves as shift register and through which a transferred bit position passes at the beginning of a transfer cycle of mark data, so that its inputs consecutively carry an output signal with which the column counter and the theoretical number register can be consecutively activated for the transfer of output signals supplied from the control store, and with which finally the control address counter is held at the address for the control word containing the first accumulator store address until the first items of mark data assigned to the transfer cycle arrive.

14. An apparatus according to any one of claims 10 to 13, in which the buffer store comprises three sub-stores each comprising four shift registers, and further comprises an input circuit for these sub-stores comprising two OR gates which are supplied in serial fashion with the mark data of the first and third and the second and third signal conversion units respectively, and which are each connected to a serial-to-parallel converter which consists of a 6-bit shift register having parallel outputs, and of which four outputs are in each case connected to the inputs of the first and third sub-stores, and the two remaining outputs are connected to the inputs of the second sub-store, and an output circuit comprising two parallel-to-serial converters which are connected in analogous fashion to the outputs of the sub-stores and whose serial outputs, for the conversion of the coded mark data into the original form, are commonly connected to an AND gate and to a further OR gate whose outputs each emit the intermediately stored mark data of the third and first signal conversion units, whereas the output of the second parallel-to-serial converter directly emits the mark data of the second signal conversion unit.

15. An apparatus for reading documents, substantially as herein described with reference to the accompanying drawings.

16. A method for the automatically recognising handwritten marks on a machine-analysable document, wherein mark boxes printed in a "drop-out" colour are optically scanned column-wise, video signals are formed in an opto-electrical transducer from which are derived digitalised items of scanning data which are to be further processed in serial fashion and which can be clearly locally assigned to the mark matrix, from which a mark statement is obtained by combining a plurality of scanning elements, in which the serially occurring video signals are fed to a plurality of parallel signal conversion units where they are quantised with stepped threshold values to form items of scanning data and these are further-processed in parallel and prepared to form items of image data, that in a scanning pattern, in which the mark boxes extend horizontally over a plurality of scanning columns and vertically over a plurality of scanning rows, the image data for each mark position is added row-wise in the column zone which is covered by the mark box so that with increasing distance from the centre of the mark box the individual image data elements are evaluated in stepped fashion with lower weighting that a plurality of row sums which follow one another in the column direction are in each case multiply combined to form image element combinations in such manner that zones of vertically adjacent mark boxes are also covered and a statement concerning a mark is derived from these determined image element combinations, that these items of mark data which have been formed in this way from the image data are intermediately stored in a buffer store and, in parallel fashion by adding the marks, established in each mark section, within the same quantisation stage and carrying out a comparison with a theoretical value, stored in a register, for the number of marks per mark section it is determined which of the read-out results of the various quantisation stages appears plausible, and that this read-out result is selected from the intermediately stored mark data and output by way of result.

17. A method according to claim 2, in which the adder means the formation of row sums black elements of the image data are added, whereas white elements located between black elements are subtracted provided a minimum value of "0" is not thereby undershot in a row sum.

18. A method according to claim 16 or 17, in which a plurality of different types of document are to be processed, each of which can possess mark sections of differeing format and a varying theoretical number of marks, in the scanning of a document firstly a code column which precedes the marking matrix in the row direction is detected and the code marks printed at that points are scanned, and the associated video signals are quantised with a high threshold and are analysed in the same way as the marks, that items of code data derived therefrom are transferred to a plausibility control unit which, in dependence thereupon, controls the sequence of the checking of items of mark data, in respect of a plausible read-out result specifically for the type of document in question.