JPH0273478A - Retrieving method for plural 2-dimensional patterns - Google Patents

Abstract

PURPOSE:To simultaneously retrieve plural 2-dimensional patterns with high efficiency by providing the state setting circuits for row and column patterns, etc., respectively, decomposing a gathering of input patterns in the row and column directions and replacing them with the state transition. CONSTITUTION:A row pattern state setting circuit 8 decides a row pattern symbol from an input gathering of patterns and at the same time sets the state transition method of a row pattern state deciding circuit 2. While a column pattern state setting circuit 9 sets the state transition method of a column pattern state deciding circuit 3 based on the row pattern symbol of the output of the circuit 8. Then a gathering of 2-dimensional patterns to be retrieved is received and decomposed into the row and column directions via the circuit 8 and a column direction pointer deciding circuit 7. These decomposed pattern gatherings are replaced with the state transition via the circuits 2 and 3, and the shift value of a pointer is previously calculated for a case where the collation of patterns fails. Thus the elements are collated with each other from the right to the left and from the bottom to the top. As a result, plural 2-dimensional patterns can be simultaneously retrieved with high efficiency.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a method for simultaneously searching for a plurality of coded two-dimensional patterns, for example, in a coded image.

(Summary of the Invention) The present invention provides a method for obtaining a plurality of data from two-dimensionally arranged data.
In the method of searching for dimensional patterns from the upper left corner of the data to the lower right corner, we align the multiple 2-dimensional patterns to be searched at the lower right corner and match them with the corresponding data from the right edge of the bottom row to the left. If matching fails midway through, the next pattern candidate to be searched is determined based on the situation, and the search target is switched based on this, thereby reducing the number of pattern matchings and shortening multiple 2D patterns at the same time. It allows you to search by time.

(Prior Art) This type of technology, for example, considers an image encoded as data and another encoded image as a pattern, and determines whether or not the data includes the pattern as a part thereof. If so, its position should be detected.

Bir is an efficient way to search for 2D patterns.
d (Bird) method, such as the literature R, S, Bird'
Two Dimensional Pattern M
"Atching" Information Process
ssing LettersIVOL, 6+Nα5+
October 1977, pp, 168-170 is known. This is Ah, which is famous for searching one-dimensional patterns.
o-Corasick method 2
This method is extended to dimensional patterns. This is done by dividing the two-dimensional pattern you want to search in the row direction, making it a row pattern, representing each with one code, and arranging them in the column direction as a column pattern. This is a method of creating a finite automaton and scanning it once on two-dimensional data from left to right and top to bottom to detect every position where a pattern appears.

With this method, it is possible to search for a plurality of two-dimensional patterns on the order of the number of data elements.

Also, when searching for multiple one-dimensional patterns, Ah.

As an efficient method other than the Corasick method, there is the method of Uratani, for example, the document Noriyoshi Uratani, "Multiple string matching algorithm", Institute of Electronics, Information and Communication Engineers, 5S87-27. There is also Japanese Patent Application No. 62-231091 entitled "Multiple Pattern Search Method" filed by the applicant. In this method, matching of the pattern group proceeds from the right end to the left, and if matching fails midway,
This method moves the point of interest on the data to the right based on a table calculated in advance from a group of patterns, and then restarts the next matching from the right end of the group of patterns. However, no method has yet been proposed that uses this method to search for multiple two-dimensional patterns.

(Problems to be Solved by the Invention) As mentioned in the section of the prior art, the conventional Bird method has the disadvantage that the search time is too long in proportion to the number of data elements. An object of the present invention is to provide a multiple two-dimensional pattern search method that incorporates the idea of Uratani's method and can efficiently search for multiple two-dimensional patterns at the same time.

(Means for Solving the Problems) In order to achieve this object, the method for searching for multiple two-dimensional patterns of the present invention is such that when a set of two-dimensional patterns to be searched is given, In the method of searching for the presence or absence of those patterns, the set of patterns is decomposed into columns and rows, each of them is replaced with a state transition, and the amount of movement of the pointer in the case of failure in pattern matching is calculated in advance, This method is characterized by searching for multiple two-dimensional patterns at the same time by matching elements in data from right to left and from bottom to top.

(Examples) The present invention will be described in detail below by way of examples with reference to the accompanying drawings.

FIG. 1 shows a schematic block diagram of the structure necessary to carry out the present invention.

In FIG. 1, a data buffer 1 writes and holds data to be searched. The row pattern state determining circuit 2 selects the elements allIQ read from the data buffer l corresponding to the pointer values p and q determined by the row pointer determining circuit 6 and the column pointer determining circuit 7 from the previous state. This is a circuit that determines the next state, and is shown in Fig. 2(a).
This is to determine the value of rst in the flowchart shown in (b). The column pattern state determination circuit 3 determines the presence or absence of a row pattern from the output rst of the row pattern state determination circuit 2, and stores it in the row pattern detection result holding buffer 5 corresponding to the array map in the flowchart shown in FIG. At the same time, the next state, that is, the value of cst in the flowchart shown in FIG. 2, is determined from rst and the data already stored in the row pattern detection result holding buffer 5.

The output determination circuit 4 receives the output r of the column pattern state determination circuit 3.
The presence or absence of output is determined from st, and if there is, it is output. The other outputs of the row pattern state determination circuit 2 and the column pattern state determination circuit 3 determine the next pointer increment -$iit, and are determined by the row direction pointer determination circuit 6 and the column direction pointer determination circuit 7, respectively. The actual value of the next pointer is determined. The row pattern state setting circuit 8 is a circuit that determines a row pattern symbol from the pattern and also sets the state transition method of the row pattern state determination circuit 2. The column pattern state setting circuit 9 is a circuit that sets the state transition method of the column pattern state determining circuit 3 from the row pattern symbol, which is one of the outputs of the row pattern state setting circuit 8.

2(a) and 2(b) are flowcharts showing the operating procedure from the data buffer 1 to the column direction pointer determination circuit 7 in the embodiment of FIG. 1, and show the procedure for searching for a pattern from data. There is. This will be explained below.

In Figure 2, rst and cst are row pattern search, respectively.
In the current state for column pattern search, p and q indicate row and column pointers that determine the data read position. The array rL Cg is for indicating the next state (when positive) or the increment of the pointer (when negative) for row pattern search and column pattern search, respectively. The array ro indicates the result of the row pattern search, and outputs a row pattern symbol when positive. The array co shows the results of the column pattern search, and when it is not empty, is used to determine the searched pattern, that is, the two-dimensional pattern to be output.

In steps St and S2 of FIG. 2(a), cst and q are initialized. S3 is a step of determining whether the data has been searched to the end. 84 to S7 are steps for initializing mm, rst, clo++p for row pattern search, and cmax, cmin,
r+*in indicates the maximum column pattern length, minimum column pattern length, and minimum row pattern length, respectively. S8 is a step of determining whether the line pattern search for the target line has been completed. If p is larger than the data row length m, the search for that row has ended, so the next row to be searched is determined in S18, and the value of +map is rearranged and initialized accordingly in S19. and return to S3.

In the row pattern search, it is determined in S9 whether the array rg is positive or negative, and if it is negative, the value of p is increased by its absolute value in S16, rst is reset to 0 in S17, and the process returns to S8. If rg is positive, the next state for the row pattern search is determined in SIO, the row pattern detection result is stored in the array map in Sll, and the state cst for the column pattern search is determined in S12.

In S13, it is determined whether cst is positive or negative, and if it is negative, S
14, dIIl is updated, the pointer is decremented by 1 in S15, and the process returns to S8. If csL is positive, Figure 2 (b
), and a column pattern search is performed in S20 to S40.

The length of the line pattern searched at 520 in Figure 2 et al.
en, the temporary column pointer S is changed to column pointer q in S21, and the pointer mc for map is changed to + in S22.
Initialized to nm. S23 is a step for determining whether or not the column pattern search has been completed; if it has been completed, the process proceeds to S15.
Return to

S24 is a step of determining whether or not there is a column pattern, that is, whether there is a two-dimensional pattern.
The value P of the row pointer and the value S of the column pointer are searched in 2.
A dimensional pattern is output. +wc at S26゜S27
and S are decremented by 1, and the data in the array map is examined in S28. If the row pattern search results are already stored in the array a+ap, the state cst for the next column pattern search is determined in S39.

S40 is a step of determining whether cst is positive or negative; if it is not negative, the process returns to S23; if it is negative, the value of dm is updated in S41, and the process returns to S15. S if the array map is “unsearched”
A partial row pattern search is performed at 29-338 to determine that value.

S29 is a dll in case the column pattern search fails.
In the step of updating , the UMC indicates elements for which the column pattern search always fails. S30 is a step of initializing a temporary row pointer j in order to partially perform a row pattern search. S31 is initialization of the state for partial row pattern search. S32 is a step of determining whether or not the row pattern search has been completed, and if it has been completed, S23 is performed.
Return to Otherwise, in S33, the value a3 of the data element is
0, and updates tst from the previous state. tst in S34
It is determined whether the value is positive or negative, and if it is negative, the process returns to S15. If not, it is determined in S35 whether a row pattern is detected at the position of the row pointer p. If a row pattern is detected then the state cs for the column pattern search at 536
t is determined and the process goes to S40 for column pattern search.

If not, j is decremented by 1 in S37, and it is determined in 338 whether the partial row pattern search has been completed;
If the process is finished, the process returns to S15; otherwise, the process returns to S32.

In normal case, 316 from the bus of execution of 510 to S15
, it can be expected that there will be more execution paths in S17, and even if it advances to 520, it will rarely go through the loop from 523 to 540, and it can be expected that it will often exit to 541 immediately at S40. The number of verifications is significantly reduced.

Figures 3 (a), (b), and (C) are Patent Application No. 62-2
No. 31091 ``Multiple Pattern Search Method'' Figures 3 (a) and (b) of the attached drawings.

(C), and the row pattern state setting circuit 8 and the column pattern state setting circuit 9 in the embodiment of FIG.
2 is a series of flowcharts showing the operating procedure for determining the values of the array 'g+cg+ro, co in FIG. 2. This will be explained below.

In Figure 3, array g is added to rg or cg in Figure 2, and array o
ut corresponds to ro or co in FIG. Further, yj indicates a row pattern when determining the array ro, and indicates a column pattern when determining the array co.

Therefore, a4. which indicates the i-th element of yJ. , is the value of the element of the pattern if yJ is a row pattern, y.

If is a column pattern, it indicates a row pattern symbol. k
is the number of row or column patterns.

The array depth indicates the number of search elements up to that point in a specific state, and the array emap indicates the state number for the last element of each pattern to be searched. Array ddm and array f indicate auxiliary functions for determining g.

It is assumed that these arrays are initialized to 0 or empty. S42 and S43 in FIG. 3(a) are initialization of new state numbers and pattern numbers. S44 is the initialization of the state number. The next state number and array dep for each element in order from the rightmost element of the pattern to the left between S45 and S55
The value of th is calculated. In S56, the array outO value is determined. In S57, the array emapO value is determined, and S
At step 59, it is determined whether all patterns have been checked.

560-579 in Part 3) are for finding the values of the array f indicating which state a certain state corresponds to before it, and also partially updating the array g. 560
~S62 is the initialization of queue and f.

A queue is a stack for holding the state to be checked. If the queue is empty in S63, see Figure 3 (
Control is transferred to C). In steps 564 to 575, the values of the array f are calculated for each state found in FIG. 3(a).

In S76 to S578, the increment of the pointer in the state fst for the element C for which the next state number S is being determined in the state r is calculated.

S80 in FIG. 3(C) is the initialization of the array dds, and S80 is the initialization of the array dds.
81 is the initialization of the pattern number. In S82 to S84, the values of bst and j ten to be used in S85 and thereafter are determined. S85 indicates that when the search fails in a certain state, if there is a corresponding state before it, the operation moves to 386 or less, and if there is no corresponding state, the operation moves to the next pattern (S95). S86 is the initialization of the queue. S87~
At S93, add a new candidate to the queue and use dda+
The value of is being updated. In S96, it is determined that the above operations for all patterns have been completed (j>k
), and in S97, the value of g at the time of verification failure is updated using the ddmO value obtained so far.

One example will be described below using a specific pattern shown in FIG. The pattern shown in Figure 4 is broken down into rows and each is called a row pattern and given a row pattern symbol. For example, A for the row pattern ""abc '" and B for "efg".
, “abcd” to C1+* cdh g-+ to D, “h
Add the symbol E to "ebc".The arrangement of these patterns in the column direction, that is, "ABAB'" and "CDE" is called a column pattern.The procedure shown in FIG. 3 is used to create an array rg.

cg, ro, and co are required. Examples of the array fgO thus obtained are shown in Table 1, and examples of the array cgO are shown in Table 2.

The set of elements here is (a, b, c, d,
e, fg, h, o) and (A, B, C,
D, E, O). Array ro is ro(3) ='
A', ro(6)='B', ro(10)='C'
, ro(13) ='D', ro(15) ='E
', and for other arguments (books) ro(-)
= “0”. Here, O'' indicates that no row pattern symbol is detected. Array co is co(4)
= ``ABAB'', co(7) ``CDE'' and the others are empty.

Table 2 An example of an embodiment of an array CG according to the present invention The search procedure shown in Figure 2 is to first search for row patterns, and when the row pattern is found, perform a column pattern search for row pattern symbols. , detection of a column pattern directly indicates detection of a two-dimensional pattern. An outline of the operation when the above method is applied to the data shown in FIG. 5 will be shown.

Since both rmin and cmin in Figure 2 are 3, the first value is 39.
The element to be matched is "f" in the third row and third column.

Then, from Table 1, the array rg becomes -1, so the row pointer is incremented by 1 in S16 of FIG. 2(a), and the process returns to S8 to proceed to the next collation. Then, since g° becomes the matching target, this time r(H becomes 4, and S10 and subsequent steps are executed.

The state of row pattern search becomes 4 in SIO, and the array map is updated in SLL. In state 4, no row pattern is detected, so ro (rs t) in 512 becomes O and cst is set to -3. Therefore, in S14, d
The value of lll remains unchanged, the row pointer p is decremented by 1 in S15, and the process returns to S8.

As the process continues, the row pattern "efg" is found in the third row and second column, so ro(rst) in S12, that is, the row pattern symbol becomes B and cst becomes 1 from Table 2,
Enter column pattern search. III at 328 in Figure 2(b)
Since ap is "unsearched °", a partial row pattern search is performed for the second row using the temporary column pointer S from S30 onwards. Continuing the process, the line pattern “a” is displayed at 535.
bc” is found, the row pattern symbol becomes A, the column pattern status cst changes to 2 in S36, and after 540, S
Return to 23.

The next line (first line) is also “unsearched, so d in S29.
After updating ll to 2, a partial row pattern search below 330 is performed. This line pattern search fails and the second
Returning to 315 in Figure (a). Here, the row pattern search for the third row is continued again. Since no more row patterns are found in this row, the process moves to 318 soon. Here, dm is 2, so the column pointer is incremented by 2. In S19, each element of the map is rearranged in preparation for the next search.

The target of the row pattern search restarted from S3 is the fifth row. Since the row pattern "efg" (row pattern symbol is B) is also found in this row, the column pattern search is performed in the same manner as described above. However, since the result of the row pattern search for the third row is stored in the array map, there is no need to perform a partial row pattern search for this row, and it is sufficient to simply use the value in S39.

As the process continues, the column pattern "ABAB" is found in S24, so a two-dimensional pattern is detected.

Overall, in the case of the data shown in FIG. 5, one two-dimensional pattern is detected in the third row, second column, and another two-dimensional pattern is detected in the sixth row, seventh column. The number of matchings in this case is 57 times in the row pattern search and 28 times in the column pattern search, for a total of 85 times, which is less than the 108 times (same as the number of elements) of the B'rrd method. In actual data, patterns usually appear less frequently, so it can be expected that this method will be even more efficient than the old RD method.

In the above explanation, the status is US! Although the movement and pointer increase/decrease are shown using the same two-dimensional arrays rg and cg that take state numbers and elements as arguments, these may be represented in separate arrays, or the values of arrays and lists can be expressed as (elements, It may also be expressed as a one-dimensional array or list by creating a set of (next state number). Furthermore, depending on the purpose of the search, even if one pattern is found as described above, the search may be stopped at this point without searching for the next pattern.

(Effect of the invention) As described above, when a set of two-dimensional patterns is given, each row of the pattern is regarded as a one-dimensional pattern, a unique symbol is assigned to each row pattern, and the symbols are arranged in the column direction. In the first step of the search, the row pattern is searched in the row direction using Uratani's method, and once the row pattern is found, the column pattern is searched in the column direction using Uratani's method. multiple two-dimensional patterns can be searched simultaneously and efficiently.

[Brief explanation of the drawing]

FIG. 1 shows a schematic block diagram of a device necessary to carry out the search method of the present invention, and FIG. FIGS. 3(a), 3(b), and 3(C) show a series of flowcharts showing the operating procedures of the row pattern state setting circuit 8 and the column pattern state setting circuit 9. , FIG. 4 shows an example of a two-dimensional pattern, and FIG. 5 shows an example of two-dimensional data. Fig. 4 a of l madan 0-ri・1 Lbc αbCd efchiJl avIr-1yty> -stop+ b)chic d o )d b f ahhcLbce9
-(10hdOe def%0hc1bdiLfe.

Claims (1)

[Claims] 1. Given a set of two-dimensional patterns to be searched,
A method for searching for the presence or absence of such patterns in two-dimensionally arranged data, the set of patterns is decomposed into columns and rows, each of which is replaced with a state transition,
Search for multiple two-dimensional patterns at the same time by calculating in advance the amount of movement of the pointer in case pattern matching fails and matching elements in the data from right to left and from bottom to top. A search method for multiple two-dimensional patterns, characterized by the following.