JP2015095042A - Image recognition device, image recognition method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an image recognition device, an image recognition method, and a program which are capable of extracting a corresponding pattern without using a mask pattern when a masked pattern is preserved.SOLUTION: The present invention, in an associative memory model in which a coupled pattern composed of a pattern-to-be-masked that is an exclusive-OR or exclusive-NOR of an original pattern of a plurality of images and a mask pattern and the mask pattern correlated thereto is learned by a neural network comprising neuron units, introduces a collation pattern that is a clue image into the associative memory model, and, by repeating a state update while decoding a state and obtaining a reconstitution pattern, compares the reconstitution pattern decoded as a result of the state update and the collation pattern when obtaining a reconstitution pattern that is the result of recollection, and imposes restriction on a neuron unit corresponding to a portion of the pattern in which values match between the reconstitution pattern and the collation pattern so that the same value as in the previous state before the update may be assumed at the time of state updating.

Description

  The present invention relates to an image recognition apparatus, an image recognition method, and a program.

  Conventionally, self-associative memory models, sequence associative memory models, and the like have been developed as associative memory models in neural networks.

  For example, in Patent Document 1, a self-propagating neural network is used in the storage layer to form a subnetwork in which nodes corresponding to the pattern of the input vector are connected, and learning is performed as a subclass of the input vector. An association is established between the key class of the key vector that is the key of the association held in the storage layer and the association class that is recalled from the key vector, and the input key vector is changed to the key class or association class. It is disclosed that the weight of a corresponding node is output as a recall result.

  In Non-Patent Document 1, a color image is converted into binary data, a bit is inverted, a reversible code is generated, and stored in a neural network, so that it is inverted with a bit expression when decoded. It is disclosed to adjust the representation to be the same integer value.

JP 2011-86132 A

“Oku, Makito, Aihara, Kazuyuki,“ Associative dynamics of color images in a large-scale I l e and p e n e i n e i n e t e n e i n e i n e i n e i t e n e i t e n e i t e n e i ” 508-521 (2011).

  However, in the conventional associative memory model, when an image pattern masked with a mask pattern is stored at the memorizing stage, it becomes a clue without using the mask pattern used for memorizing at the recall stage. There is a problem that the recall pattern cannot be obtained only from the matching pattern.

  The present invention has been made in view of the above problems, and in an associative memory model neural network in which a masked pattern is stored, a corresponding pattern can be extracted without using a mask pattern. It is an object to provide an apparatus, an image recognition method, and a program.

  In order to achieve such an object, an image recognition device of the present invention is an image recognition device including at least a storage unit and a control unit, and the storage unit excludes a plurality of original patterns and mask patterns. Information relating to an associative memory model in which a neural network composed of neural cell units learns a connected pattern in which the mask pattern is associated with a masked pattern that is a logical OR or a negative exclusive OR, and the control unit A recollection means for obtaining a reconstructed pattern of a recollection result by introducing a specified collation pattern, which is a clue image, into the associative memory model, decoding a state and obtaining a reconstructed pattern and repeating state update; The reconstructed pattern decrypted with the state update by the recall means is compared with the collation pattern, For the above-mentioned neuronal cell unit corresponding to the portion where the value matches between the configuration pattern and the matching pattern, it is constrained to easily take the same value as the state before the update when the state is updated, while the value matches. The nerve cell unit corresponding to the portion not to be processed is characterized by comprising comparison means for constraining the state update so as to easily take a value different from the state before the update.

  Further, the image recognition device of the present invention is the image recognition device described above, wherein the recall means converges the state of the neural network by gradually lowering the temperature with the state update by annealing. It is characterized by controlling so that the value of the energy function does not fall into a local minimum.

  The image recognition apparatus according to the present invention is the image recognition apparatus described above, wherein the recalling means exponentially lowers the temperature by multiplying the inverse temperature by an increase rate for each state update. It is characterized by.

  Further, the image recognition apparatus according to the present invention is the image recognition apparatus described above, wherein the comparison unit adds the restriction according to a sum of products of values between the reconstruction pattern and the collation pattern. It is characterized by setting a constraint term that reduces the value of the energy function of the neural network.

  The image recognition device according to the present invention is the image recognition device described above, wherein the internal potential of the nerve cell unit is obtained by inverting the sign by partially differentiating the energy function with the constraint term in the state of the nerve cell unit. The state update probability of the neuronal unit is the index obtained by adding 1 to the base obtained by inverting the sign of the product of the internal potential and the inverse temperature, and 1 is the numerator. The recall means recalculates the state update probability every time the state is updated.

  The image recognition apparatus according to the present invention is the image recognition apparatus described above, wherein the recall means sets the neural network to a random value in an initial state, and follows the neural network according to a synchronous and probabilistic update rule. Updating the state of.

  The image recognition apparatus of the present invention is characterized in that, in the image recognition apparatus described above, the mask pattern is a random binary pattern.

  In the image recognition device according to the present invention, in the image recognition device described above, the mask pattern is generated by the mask pattern common to any number of bits in the original pattern among the plurality of images. It is characterized by being a pattern.

  The present invention also relates to an image recognition method. The image recognition method of the present invention is an image recognition method executed in a computer including at least a storage unit and a control unit, and the storage unit includes a plurality of images. An associative memory model in which a neural network composed of neural cell units learns a connected pattern in which a mask pattern is associated with a masked pattern that is an exclusive OR or a negative exclusive OR of an original pattern and a mask pattern Information is stored, and a matching pattern that is a designated clue image that is executed by the control unit is introduced into the associative memory model, and state update is repeated while obtaining a reconstruction pattern by decoding the state. With the recall procedure for obtaining the reconstructed pattern of recall results and the state update by the recall means. The decoded reconstructed pattern is compared with the collation pattern, and the neural cell unit corresponding to the part where the value matches between the reconstructed pattern and the collation pattern is updated when the state is updated. Constraining the same value as the previous state to be easy, while restricting the neuron unit corresponding to the portion where the value does not match to easily take a value different from the state before the update when the state is updated. And a comparison procedure to be added.

  The present invention also relates to a program. The program of the present invention is a program for causing a computer including at least a storage unit and a control unit to execute the program. The storage unit includes a plurality of image original patterns and masks. Storing information on an associative memory model in which a neural network composed of nerve cell units learns a connected pattern in which the mask pattern is associated with a masked pattern that is exclusive OR or negative exclusive OR with a pattern; Then, by introducing the collation pattern, which is the designated cue image, executed in the control unit into the associative memory model, and decoding the state to obtain the reconstructed pattern and repeating the state update, the recall result Recall procedure for reconstructing pattern and decoding with the above state update by the above recall means The reconstructed pattern is compared with the collation pattern, and the neural cell unit corresponding to the portion where the value is identical between the reconstructed pattern and the collation pattern is not updated at the time of the state update. A comparison that restricts the same value as the state to be easily obtained, and restricts the nerve cell unit corresponding to the portion where the value does not match to restrict a value that is different from the state before the update when the state is updated. And a procedure is executed.

  The present invention also relates to a recording medium, wherein the program described above is recorded.

  According to this invention, a neural network comprising a neuron unit as a connected pattern in which a mask pattern is associated with a mask pattern that is an exclusive OR or a negative exclusive OR of an original pattern and a mask pattern of a plurality of images. In the associative memory model learned in the above, when the reconstitution pattern of the recall result is obtained by introducing the collation pattern, which is a cue image, into the associative memory model and repeating the state update while decoding the state and obtaining the reconstructed pattern In addition, the reconstructed pattern decoded with the state update is compared with the collation pattern, and the state update is performed for the neuronal cell unit corresponding to the portion where the value matches between the reconstructed pattern and the collation pattern. Sometimes it is constrained so that the same value as that before the update can be easily obtained. Thereby, in the associative memory model, when a masked pattern is stored, the corresponding pattern can be extracted without using the mask pattern.

  In the present invention, the state of the neural network is converged by gradually lowering the temperature with the state update by the annealing method in the recall stage, so that the energy function value is controlled not to fall into the local minimum. There is an effect that can be.

  In addition, since the present invention exponentially lowers the temperature by multiplying the inverse temperature by the rate of increase every time the state is updated once, in the annealing method, the state of the nerve cell unit is initially changed. As a result, the temperature can be gradually lowered and converged.

  Further, according to the present invention, in order to add a constraint, a constraint term that reduces the value of the energy function of the neural network is set in accordance with the sum of products of values between the reconstruction pattern and the matching pattern. As a result, the present invention has an effect that a bias can be applied so that the energy function decreases as the values match between the reconstruction pattern and the matching pattern.

  In the present invention, the internal potential of the nerve cell unit is obtained by partially differentiating the energy function with a constraint term in the state of the nerve cell unit and inverting the sign, and the state update probability of the nerve cell unit is: An index with the sign of the product of the internal potential and the inverse temperature inverted, plus 1 plus the denominator, and expressed as a fraction with 1 as the numerator. Then, the state update probability is recalculated. As a result, while the temperature is high or when the values between the reconstruction pattern and the matching pattern do not match, the state of the nerve cell unit is changed and the temperature gradually decreases. When the values coincide with each other, there is an effect that it can be applied so as to converge to a certain state.

  In addition, the present invention sets the neural network to a random value in the initial state, and updates the state of the neural network according to a synchronous and probabilistic update rule, so that the corresponding pattern can be suitably recalled. There is an effect.

  In the present invention, since the mask pattern is a random binary pattern, it is possible to find a pattern corresponding to the collation pattern from a pseudo-orthogonalized neural network having a large storage capacity.

  In the present invention, since the mask pattern is a pattern created by a mask pattern common to any number of bits in the original pattern of a plurality of images, the connection pattern can be extended by increasing the mask pattern. There is an effect that a pattern corresponding to the matching pattern can be found from the suppressed neural network having a large storage capacity.

FIG. 1 is a diagram schematically showing the memorizing stage of the associative memory model. FIG. 2 is a diagram schematically showing the recall stage of the associative memory model. FIG. 3 is a diagram schematically showing a case where patterns having a small correlation with each other are stored. FIG. 4 is a diagram schematically showing a case where patterns having a large correlation with each other are stored. FIG. 5 is a diagram illustrating a relationship among the original pattern, the mask pattern, the mask pattern, and the connection pattern. FIG. 6 is a diagram illustrating an example of the XNOR operation and the XOR operation. FIG. 7 is a diagram illustrating an example of a connection pattern. Figure 8 is a diagram showing an example of connection pattern of the specified mask pattern r _k different for each block. FIG. 9 is a block diagram illustrating an example of the image recognition apparatus 100 to which the exemplary embodiment is applied. FIG. 10 is a flowchart showing an example of basic processing of the image recognition apparatus 100 in the present embodiment. FIG. 11 is a flowchart showing an example of the recall process of the image recognition apparatus 100 in the present embodiment. FIG. 12 is a diagram showing the results of image recall without using mask information according to this example.

  Embodiments of an image recognition apparatus, an image recognition method, and a program according to the present invention will be described below in detail with reference to the drawings. Note that the present invention is not limited to the embodiments. For example, the present invention may be applied to various associative memory models as well as the associative memory model shown in the present embodiment. In the following embodiments, an example in which synchronous and probabilistic updating is performed may be described. However, the present invention is not limited to this, and asynchronous updating or deterministic updating may be performed. In this embodiment, the case where the neural network is loosely coupled may be described as an example. However, the present invention is not limited to this, and the neural network is tightly coupled (for example, all the neural cell units are coupled to each other). Network). In the following embodiments, a mask pattern may be described by taking a random binary pattern as an example. However, the present invention is not limited to this, and the present embodiment can be applied to any mask pattern.

[Background of the present embodiment]
Hereinafter, the background and outline of the embodiment of the present invention will be described with reference to FIGS. 1 to 8, and then the configuration and processing of the present embodiment will be described in detail.

  First, in the embodiment of the present invention, as a premise, since the masked pattern is stored in the neural network of the associative memory model, the background will be described first. Here, FIG. 1 is a diagram schematically showing the memorizing stage of the associative memory model.

  An associative memory model is a kind of neural network model. The Hopfield network, which is a type of associative memory model, is a model of a neural network proposed by Hopfield in 1982. As an example, as shown in FIG. 1, a neural network in which nerve cell (neuron) units are coupled is constructed and interacts between nerve cell units. The strength of connection between nerve cell units is expressed as a weight (connection coefficient), and the strength increases or decreases by learning.

  FIG. 1 schematically shows an inscription stage for storing a binary image of four patterns in a neural network. In the memorization stage of the self-associative memory model, a plurality of memory patterns are stored, and in the recall stage, one pattern is recalled. Whether or not each neuron unit in the network is fired is determined by the sum of inputs from other neuron units to be connected. The memory pattern in the auto-associative memory model is a way of arranging the activity states of neuronal cell units (typically, two types of firing state and resting state). In the Hopfield model, the memory pattern is stored in the neural network by determining the weight of the synapse connection according to a specific learning rule (such as the Hebb rule) at the inscription stage. Further, at the recall stage, the energy function representing the network state monotonously decreases according to the state update rule, and the update process is repeatedly executed until a minimum value state is reached. FIG. 2 is a diagram schematically showing the recall stage of the associative memory model.

  As shown in FIG. 2, in the recall stage of the self-associative memory model, the original pattern stored in the neural network is searched from the collation pattern serving as a clue to obtain the recalled pattern. More specifically, when retrieving a memory pattern, first, an initial pattern as a clue is given to the neural network, and then a desired memory pattern is updated by updating the state of the neural network a plurality of times according to a specific rule. Get. Note that the clue pattern does not have to be the original pattern itself, but may be a part of the pattern desired to be obtained. That is, in the self-associative memory model, a stored pattern that is most similar to the cue pattern given as an input is obtained.

  The number of patterns that can be stored increases as the number of neurons (number of neuronal cell units) N in the network increases, but the number of patterns that can be stored is limited by the correlation between patterns. The number of patterns in which the self-associative memory model can be remembered is called memory capacity. It is known that when a standard learning rule or state update rule is used as the upper limit of the theoretical storage capacity, it is about 0.14 N with respect to the size N of the neural network and the storage pattern. However, in reality, it is greatly influenced by the correlation of the storage patterns. Here, FIG. 3 is a diagram schematically showing a case where patterns having a small correlation are stored, and FIG. 4 is a diagram schematically showing a case where patterns having a high correlation are stored.

  As shown in FIG. 3, when the storage pattern is a random pattern and the correlation is small, a storage capacity of 0.14N close to the theoretical limit can be obtained. However, as shown in FIG. 4, when patterns similar to each other are stored, since the correlation between the storage patterns is large, they generally interfere with each other and the storage capacity generally decreases significantly.

  Conventionally, (1) a method of improving a learning rule by using a pseudo inverse matrix or sequentially adjusting weights, and (2) adding a bit to the end or a bit against the problem of a decrease in storage capacity. A method of changing a pattern itself by changing a part of the pattern has been proposed. However, since all of the conventional methods have a large calculation load, it is difficult to apply when the size N of the neural network and the storage pattern is large.

  Therefore, the inventors of the present application have developed a new method through intensive studies. As an example, the new method of memorization used in this application is (1) preparing a mask pattern for a memory pattern (original pattern), and (2) applying a mask pattern by XNOR or XOR operation of the memory pattern and the mask pattern. The basic element is to generate and (3) store a connected pattern obtained by connecting a mask pattern and a mask pattern in a neural network. In this way, by using the mask pattern for randomization, the pattern to be stored can be pseudo-orthogonalized, and the reduction in storage capacity due to correlation can be suppressed. Here, FIG. 5 is a diagram illustrating the relationship among the original pattern, the mask pattern, the mask pattern, and the connection pattern.

  An XOR operation or XNOR operation is performed for the purpose of randomizing the binary signal sequence. As shown in FIG. 5, as a principle, an original signal sequence (original pattern) and a (pseudo) random number sequence (random mask) for masking it are prepared, and XOR operation or XNOR operation of both is performed. Thus, a messy signal sequence (masked pattern) is obtained. As a result, the storage pattern becomes confused and the correlation becomes small, interference between patterns can be reduced, and a reduction in storage capacity can be prevented.

  More specifically, in the embodiment of the present invention, (1) a random binary pattern (random mask) mask pattern is generated. Note that the binary appearance probability is not limited to ½ evenly, and a slight bias may be applied.

  In this embodiment, (2) a masked pattern is generated by obtaining an exclusive OR (XOR) or a negative exclusive OR (XNOR) of the original pattern of a plurality of designated images and the mask pattern. To do. As an example, the present embodiment uses an exclusive OR of values corresponding to both data using one-dimensional binary data of an image as an original pattern and random binary data of the same or short length as a mask pattern. XOR) or negative exclusive OR (XNOR) is obtained to generate a masked pattern. Here, FIG. 6 is a diagram illustrating an example of the XNOR operation and the XOR operation.

When binary data of {0, 1} is used for the original pattern (Input 1) and the mask pattern (Input 2), the XNOR operation shown in FIG. 6A is performed to obtain the mask pattern value (Output). The binary data may be any two values. For example, when binary data having an original pattern (Original) and a mask pattern (Mask) of {1, -1} is used, FIG. The XNOR operation shown in b) may be performed to obtain the masked pattern value (Masked). The XNOR calculation in this case corresponds to the product of the original pattern r _i and the mask pattern ξ _i multiplied by r _i × ξ _i .

In one XOR operation, when binary data whose original pattern (Input1) and mask pattern (Input2) are {0, 1} is used, the XOR operation shown in FIG. 6C is performed, and the value of the masked pattern (Output) ) Also in this case, the binary data may be any two values. As an example, when binary data having an original pattern (Original) and a mask pattern (Mask) of {1, -1} is used, The value of the masked pattern (Masked) may be obtained by performing the XOR operation shown in FIG. The XOR operation in this case corresponds to a value obtained by multiplying the original pattern r _i and the mask pattern ξ _i and exchanging the signs, −r _i × ξ _i .

  Subsequently, in the memorizing stage, the present embodiment (3) learns by using an associative memory model using a neural network with a mask pattern associated with the mask pattern as a connected pattern. The order of bits in the connection pattern may be any order. Moreover, the connection pattern completely holds the information of the original pattern. This is because the original pattern can be restored by performing the same logical operation (XNOR / XOR) on the mask pattern / masked pattern pair again. Here, FIG. 7 is a diagram illustrating an example of a connection pattern.

As shown in FIG. 7 as an example, in this embodiment, and the object to be mask pattern r _k xi] _k performing the XOR / XNOR operation using the mask pattern r _k different for each one of the original pattern xi] _k, may be allowed to memorization as connection pattern of the mask pattern r _k alternately arranged (k is the block number). In this case, if the length of the original pattern is N, a connected pattern having a length of 2N is generated. Therefore, the number of necessary neurons (number of nerve cell units) doubles.

Therefore, a plurality of original patterns may be divided in units of blocks, and different mask patterns (such as random masks) may be set for each block unit to reduce the length of the connection pattern. That is, the length of the connected pattern may be made shorter than 2N by preparing a smaller number of bits of the mask pattern than the original pattern and using the same element (mask) following a plurality of XNOR operations. Here, FIG. 8 is a diagram showing an example of connection pattern of the specified mask pattern r _k different for each block. As shown in FIG. 8, in this embodiment, the original pattern of a plurality of images is divided into L-1 block units (blocks of length L including a mask pattern) from the top, and a value is obtained for each block. May be set to generate a connected pattern having a length of LN / (L-1). In the case of FIG. 7, the number of necessary neurons (number of neuronal cell units) is doubled, but by setting the block length to L as shown in FIG. 8, the number can be suppressed to L / (L-1) times.

[Outline of this embodiment]
In the present embodiment, as an example, assuming that the concatenated pattern masked as described above is stored in the neural network of the associative memory model, the corresponding pattern is used from the neural network without using the mask pattern. The purpose is to take out. In the recall stage following the memorization stage, the original pattern stored in the neural network is usually searched from the input pattern as a clue to obtain the recalled pattern.However, when this embodiment is applied, The mask pattern corresponding to the verification pattern is unknown.

  Therefore, in the present embodiment, a reconstitution pattern of the recall result is obtained by introducing a collation pattern, which is a clue image, into the associative memory model, and repeating state updating while obtaining the reconstructed pattern by decoding the state.

  In this case, the present embodiment compares the reconstructed pattern decoded along with the state update with the collation pattern, and the nerve corresponding to the portion where the value between the reconstructed pattern and the collation pattern matches. For cell units, it is constrained to easily take the same value as the state before update at the time of state update. On the other hand, for nerve cell units corresponding to portions where the values do not match, values different from the state before update at the time of state update Add restrictions to make it easier to take Due to this restriction, the state of the neuron unit that was initially random tends to stay in the same state in the neuron unit corresponding to the matching part, and gradually approaches the state reflecting the unknown mask pattern as the state is updated. It will follow.

  In other words, in this embodiment, the initial state of the random neural network is optimized to a state in which the mask pattern is gradually reflected. However, if optimization is not appropriate, the global minimum reflecting the true mask pattern may not be reached, and the local minimum may be encountered.

Here, the present embodiment may use an annealing method (simulated annealing method) as an optimization method. More specifically, the present embodiment uses an annealing method to gradually lower the temperature as the state is updated, thereby converging the state of the neural network so that the energy function value does not fall into the local minimum. You may control. At that time, as indicated by the following formula, the temperature may be decreased exponentially by multiplying the reverse temperature by the increasing rate for each state update.
β (t + 1) = γβ (t)
(Where t is the time associated with the state update, β (t) is the reverse temperature at time t, and γ is the rate of increase and takes a value greater than 1.)

Further, in this embodiment, in order to add the above-described constraints, the value of the energy function of the neural network is reduced according to the sum of products of values between the reconstructed pattern and the matching pattern, as shown in the following formula. A constraint term may be set.
(Here, the first term of the energy function E is a normal energy function term, and the second term is a constraint term added in the present embodiment. J _ij is the j-th neuron. Represents the strength of the synaptic connection from the unit to the i-th neuron unit, and x _i (t) is the state of the i-th neuron unit at time t, and s is a parameter representing the strength of the constraint term Z _k is a value of a collation pattern serving as a clue, and y _k is a value of a reconstruction pattern.)

In the present embodiment, as the internal potential h _i (t) of a neuron unit, the energy function E with a constraint term is partially differentiated with the state x _i of the neuron unit as shown in the following equation. It is possible to set a reversed sign.

In the present embodiment, the state update probability Prob (x _i (t + 1) = 1) of the neuron unit is expressed by the sign of the product of the internal potential h _i (t) and the reverse temperature, as shown in the following equation. A fraction obtained by adding 1 to an index with the inverted number as the base and 1 as the numerator may be used.

  As an example, this embodiment recalculates the state update probability described above for each state update, and updates the state of the nerve cell unit according to the recalculated probability. As a result, in the initial stage where the temperature is high or the value between the reconstruction pattern and the matching pattern does not match, the state of the corresponding nerve cell unit is changed, while the temperature gradually increases. When the value decreases between the reconstructed pattern and the matching pattern, the state of each neuron unit is applied so as to converge to a certain state.

  The above is the outline of the present embodiment.

[Configuration of image recognition device]
Next, the configuration of the image recognition apparatus will be described with reference to FIG. FIG. 9 is a block diagram showing an example of the image recognition apparatus 100 to which the present embodiment is applied, and conceptually shows only the portion related to the present embodiment in the configuration.

  As shown in FIG. 9, the image recognition apparatus 100 according to the present embodiment schematically includes at least a control unit 102 and a storage unit 106. In the present embodiment, the image recognition apparatus 100 further controls communication with the input / output control interface unit 108. An interface unit 104 is provided. Here, the control unit 102 is a CPU or the like that comprehensively controls the entire image recognition apparatus 100. The communication control interface unit 104 is an interface connected to a communication device (not shown) such as a router connected to a communication line or the like, and the input / output control interface unit 108 is connected to the input unit 112 or the output unit 114. The interface to be connected. The storage unit 106 is a device that stores various databases and tables. Each unit of the image recognition apparatus 100 is connected to be communicable via an arbitrary communication path. Further, the image recognition apparatus 100 is communicably connected to the network 300 via a communication device such as a router and a wired or wireless communication line such as a dedicated line.

  Various databases and tables (neuron file 106b and the like) stored in the storage unit 106 are storage means such as a fixed disk device. For example, the storage unit 106 stores various programs, tables, files, databases, web pages, and the like used for various processes.

Among these components of the storage unit 106, the neuron file 106b relates to a neural network such as a parameter between nerve cell units and a parameter (coefficient value) for each nerve cell unit in association with an index for identifying the nerve cell unit. Neural network storage means for storing data. As an example, the neural network information stored in the neuron file 106b is the nerve at the weight J _ij between the neuron unit i and the neuron unit j at time t (t = 1, 2, 3,...). These are values such as the state x (t) of the network, the reconstruction pattern y (t), the energy function E (t), and the internal potential h (t) of the nerve cell unit. The neuron file 106b may store initial parameters when numerical values are not given in advance when these parameters t = 0. The above-mentioned value may be updated every time state update (t → t + 1) is performed by the learning unit 102c and the recall unit 102d described later.

  The input / output control interface unit 108 controls the input unit 112 and the output unit 114. Here, as the output unit 114, in addition to a monitor (including a home television), a speaker can be used (hereinafter, the output unit 114 may be described as a monitor). As the input unit 112, a keyboard, a mouse, a microphone, and the like can be used.

  In FIG. 9, the control unit 102 has an internal memory for storing a control program such as an OS (Operating System), a program defining various processing procedures, and necessary data. And the control part 102 performs the information processing for performing various processes by these programs. The control unit 102 includes a mask generation unit 102a, a logical operation unit 102b, a learning unit 102c, a recall unit 102d, and a comparison unit 102e in terms of functional concept.

Among these, the mask generation unit 102a is a mask generation unit that generates a mask pattern. In the present embodiment, the mask generation unit 102a generates a mask pattern that is a random binary pattern. As an example, the mask generation unit 102a may generate a random number sequence r ^μ (μ = 1, 2,..., P) using a random number table or the like. It should be noted that the appearance probability Prob (r _i ^μ = ± 1) of the generated binary values (for example, 1 and −1) is not limited to ½, and a slight bias (bias parameter a) is added. (1 ± a) / 2.

  The logical operation unit 102b is a logical operation unit that generates a masked pattern by performing a logical operation (XNOR / XOR) between the mask pattern (random mask or the like) generated by the mask generation unit 102a and the original pattern. is there. For example, the logical operation unit 102b performs an XNOR operation or an XOR operation as described above with reference to FIG.

  Here, the logical operation unit 102b creates a mask pattern by setting a common mask pattern (random mask, etc.) in units of blocks by making a block for each arbitrary number of bits in the original pattern of a plurality of images. May be. More specifically, the logical operation unit 102b divides an original pattern of length N by a plurality of images into L-1 blocks from the beginning (where N is divisible by L-1). ), A mask pattern having a different value for each block may be set to generate a concatenated pattern having a length of LN / (L−1).

  The learning unit 102c is learning means for learning a connection pattern using an associative memory model using a neural network.

  The recalling unit 102d is a recalling unit that reconstructs a pattern from a matching pattern that is a clue in an associative memory model using a neural network. More specifically, the recall unit 102 introduces a collation pattern, which is a clue image, into the associative memory model, and repeats state update while decoding the state and obtaining the reconstruction pattern, thereby obtaining the reconstruction pattern of the recall result. Ask. Here, the recalling unit 102d updates the state of the neural network while receiving restrictions on the state update of each nerve cell unit set by the comparison unit 102e.

  Here, the recalling unit 102d may converge the neural network using an annealing method in order to guide the unknown mask pattern to the determined state. For example, the recalling unit 102d may control the energy function value so as not to fall into a local minimum by converging the state of the neural network by gradually lowering the temperature as the state is updated. Here, the recalling unit 102d may lower the temperature exponentially by multiplying the inverse temperature by the increase rate for each state update. The recalling unit 102d may lower the temperature logarithmically. For example, the recalling unit 102d may perform asynchronous update instead of synchronous update, and lower the temperature logarithmically to facilitate the global minimum.

  In addition, the comparison unit 102e compares the reconstructed pattern decoded by the recall unit 102d with the state update with the collation pattern, and corresponds to a portion where the value matches between the reconstructed pattern and the collation pattern. For the neuron unit to be updated, it is constrained to easily take the same value as the state before update at the time of state update, while for the neuron unit corresponding to the portion where the value does not match, the state before update at the time of state update This is a comparison means for adding a restriction so that different values can be easily obtained.

Here, explanation will be given using mathematical formulas. The original pattern to be stored is {ξ ^μ ₁ , ξ ^μ ₂ ,. . . , Ξ ^μ _N }. μ is a pattern number and takes an integer value from 1 to P. ξ ^μ _i represents the i-th value of the μ-th original pattern, and takes a value of 1 or −1 in the present embodiment. On the other hand, the mask pattern is {r ^μ ₁ , r ^μ ₂ ^,. . . , R ^μ _N }. In this embodiment, r ^μ _i indicates the i-th value of the μ-th mask pattern, and takes a value of 1 or −1 at random. The result of the XNOR operation between a value ξ ^μ _i having a stored pattern and the value r ^μ _i of the corresponding mask pattern is ξ ^μ _i r ^μ _i (in the case of XOR operation, −ξ ^μ _i r ^μ _i ).

When the block length L = 2, the mask pattern {r ^μ ₁ , r ^μ ₂ ^,. . . , R ^μ _N } and the masked pattern {ξ ^μ ₁ r ^μ ₁ , ξ ^μ ₂ r ^μ ₂ ,. . . , Xi] ^mu _N one by one from r ^mu _N} that connecting fetches the alternating _{^{{r μ 1, ξ μ 1}} r μ 1, r μ 2, ξ μ 2 r μ 2,. . . , R ^μ _N , ξ ^μ _N r ^μ _N } is a connection pattern of length 2N to be stored in the neural network under the control of the learning unit 102c. Here, the symbols are replaced, and the connected patterns {η ^μ ₁ , η ^μ ₂ ,. . . , Η ^μ _{N ′} }. However, N ′ = 2N.

When the block length L = 2 or more, the k-th block of the μ-th original pattern is represented by {ξ ^μ _{k, 1} , ξ ^μ _{k, 2} ,. . . , Ξ ^μ _{k, L−1} }. The value of the mask pattern corresponding to this is set to r ^μ _k , and a logical operation (XNOR / XOR) and a concatenation operation are performed, so that a block of length L {r ^μ _k , r ^μ _k ξ ^μ _{k, 1} , r ^μ _k ξ ^μ _{k, 2} ,. . . , R ^μ _k ξ ^μ _{k, L−1} }. It is good also considering what arranged this in block order as a connection pattern. In this case, N ′ = LN / (L−1). Note that the arrangement order of the above connection patterns is an example, and the arrangement order of the mask pattern and the mask pattern in the connection pattern is arbitrary.

For example, a standard Hebb learning rule may be used for learning the connection pattern by the learning unit 102c. In the case of the auto-associative memory model, the strength of the synapse connection from the j-th neuron unit to the i-th neuron unit is represented as J _ij, and in the case of i ≠ j, the learning unit 102c calculates a value according to the following equation: Determine. If i = j, J _ii = 0.

The state of the neural network at time t is expressed as {x ₁ (t), x ₂ (t),. . . , X _{N ′} (t)}. x _i (t) represents the active state of the i-th neuron unit and takes a value of 1 (firing state) or −1 (resting state). There are four possible ways to update the state of the neural network, depending on the combination of synchronous update and asynchronous update, and deterministic update and probabilistic update. Any combination may be adopted.

  In the present embodiment, the recollection unit 102d matches the reconstructed pattern obtained by decoding the state of the neural network with the collation pattern that is a clue image as much as possible by the processing of the comparison unit 102e when the state of the neural network is updated. Recalling may be performed by gradually converging the state of the neural network by an annealing method while adding constraint terms.

  Here, the constraint term of the present embodiment reflects the comparison result between the reconstructed pattern and the collation pattern by the comparison unit 102e. Restrictions are made so that it is easy to take the same value as the current state at the next update, and for the part where the values do not match, the corresponding nerve cell unit is set to a value different from the current state at the next update. This is a term that adds a constraint so that it is easy to take. This constraint term makes it easier for the reconstruction pattern and the matching pattern to match.

In the present embodiment, the recalling unit 102d may use the following annealing method as an example. First, the recall unit 102d sets the initial state of the neural network to a random value, and sets an initial value for the inverse temperature parameter β. Subsequently, the recalling unit 102d updates the state of the neural network by one step according to the synchronous-stochastic update rule. Thereafter, the recalling unit 102d recalculates the constraint term reflecting the comparison result by the comparing unit 102e, increases the inverse temperature parameter β, and updates the state of the neural network again. The recall unit 102d may repeat the above process for a predetermined number of steps, and output the last obtained reconstruction pattern to the output unit 114 or the like. Here, the formulas are organized as follows.
Original pattern (length N): {ξ ^μ ₁ , ξ ^μ ₂ ,. . . , Ξ ^μ _N } (μ = 1, 2,..., P)
Mask pattern (length N): {r ^μ ₁ , r ^μ ₂ ^,. . . , R ^μ _N } (μ = 1, 2,..., P)
Mask pattern (length N): {ξ ^μ ₁ r ^μ ₁ , ξ ^μ ₂ r ^μ ₂ ,. . . _{^{, Ξ μ N r μ N}}} (μ = 1,2, ..., P)
Connection pattern (length 2N): {η ^μ ₁ , η ^μ ₂ ,. . . , Η ^μ _2N } (μ = 1, 2,..., P)
Matching pattern (length N): {z ₁ , z ₂ ,. . . , Z _N }
Neural network state: {x ₁ (t), x ₂ (t),. . . , X _2N (t)} (t = 0, 1, 2,...)
Reconstruction pattern (length N): {y ₁ (t), y ₂ (t),. . . , Y _N (t)} (t = 0, 1, 2,...)

As an example, the recall unit 102d may obtain the value of the energy function E with a constraint term that reflects the comparison result by the comparison unit 102e for each state update, using the following formula.

As an example, the recall unit 102d may obtain the internal potential h _i (t) of each nerve cell unit of the neural network using the following mathematical formula.

Further, as an example, the recalling unit 102d may obtain the state update probability Prob (x _i (t + 1) = 1) of each nerve cell unit of the neural network using the following formula.
Here, each time the state is updated (t → t + 1), the recalling unit 102d resets the reverse temperature β at the increase rate β (that is, β (t + 1) = γβ (t)).

As described above, as an example, the recalling unit 102d has an internal potential h _i (t) and an inverse temperature β (t by the energy function E with a constraint term reflecting the comparison result by the comparison unit 102e for each state update. ) And recalculate the state update probability of each neuronal unit defined by them. Due to the state update probability of each neuron unit that is sequentially updated, the state update probability of the corresponding neuron unit in the initial condition where the reverse temperature β is low or in the part where the value does not match between the reconstructed pattern and the matching pattern Is high, the state x (t) of the nerve cell unit is changed (0⇔1). On the other hand, when the inverse temperature β (t) rises exponentially (γ ^t−1 ) and the values match between the reconstruction pattern and the matching pattern, the state update probability of the corresponding neuron unit is low. Thus, the state of the nerve cell unit falls to a certain stable state, and the entire neural network converges.

On the other hand, the recalling unit 102d considers that the state has converged when the state is updated by a predetermined number of steps, and outputs a reconstruction pattern obtained by decoding the state at that time to the output unit 114 or the like. If the value of the energy function below the threshold cannot be obtained, it is assumed that the recall has failed, and the recall process is repeated from the initial state again after resetting the value of the reverse temperature increase rate γ, etc. May be. Note that the recalling unit 102d may decode the state of the neural network using the following equation. Note that the reconstruction pattern is expressed as {y ₁ (t), y ₂ (t),. . . , Y _N (t)}.
y _k (t) = x _2k−1 (t) x _2k (t)

  The above is an example of the configuration of the image recognition apparatus 100 in the present embodiment. Note that the image recognition apparatus 100 may be connected to the external system 200 via the network 300. In this case, the communication control interface unit 104 performs communication control between the image recognition device 100 and the network 300 (or a communication device such as a router). That is, the communication control interface unit 104 has a function of communicating data with other terminals via a communication line. The network 300 has a function of connecting the image recognition apparatus 100 and the external system 200 to each other, for example, the Internet.

  The external system 200 is mutually connected to the image recognition apparatus 100 via the network 300, an external database regarding various parameters, a program for causing the connected computer (information processing apparatus) to execute an image recognition method, and the like. It has a function to provide.

  Here, the external system 200 may be configured as a WEB server, an ASP server, or the like. Further, the hardware configuration of the external system 200 may be configured by an information processing apparatus such as a commercially available workstation or a personal computer and its attached devices. Each function of the external system 200 is realized by a CPU, a disk device, a memory device, an input device, an output device, a communication control device, and the like in the hardware configuration of the external system 200 and a program for controlling them.

  This is the end of the description of the configuration of the present embodiment.

[Processing of Image Recognition Device 100]
Next, an example of processing of the image recognition apparatus 100 according to the present embodiment configured as described above will be described in detail with reference to FIG. FIG. 10 is a flowchart showing an example of basic processing of the image recognition apparatus 100 in the present embodiment.

As shown in FIG. 10, first, the mask generation unit 102a generates a random mask that is a random binary pattern (step SA-1). As an example, the mask generation unit 102a may generate a random number sequence r ^μ (μ = 1, 2,..., P) using a random number table or the like.

  Then, the logic operation unit 102b generates a masked pattern by performing a logic operation (XNOR operation or XOR operation) between the mask pattern (random mask) generated by the mask generation unit 102a and the original pattern (step SA- 2).

  And the learning part 102c is made to learn a connection pattern by the associative memory model using a neural network (step SA-3).

  Then, the recalling unit 102d introduces a collation pattern, which is a clue image, into the associative memory model, obtains a reconstruction pattern by decoding the state and obtains a reconstruction pattern, and obtains a reconstruction pattern of the recall result (step) SA-4). At that time, the recalling unit 102d updates the state of the neural network while receiving restrictions on the state update of each nerve cell unit set for each state update by the comparison unit 102e. That is, the comparison unit 102e compares the reconstructed pattern decoded by the recalling unit 102d with the state update and the collation pattern, and corresponds to a portion where the value matches between the reconstructed pattern and the collation pattern. For the neuron unit to be updated, it is constrained to easily take the same value as the state before update at the time of state update, while for the neuron unit corresponding to the portion where the value does not match, the state before update at the time of state update Add constraints to make it easier to take different values. When the state is updated by a predetermined number of steps, the recalling unit 102d outputs the reconstructed pattern obtained by decoding the state at that time to the output unit 114 and ends the process.

  The above is an example of the basic processing of the image recognition apparatus 100 in the present embodiment.

[Recall processing]
Next, an example of a recall process that embodies part of the basic process of the image recognition apparatus 100 will be described below with reference to FIG. FIG. 11 is a flowchart showing an example of the recall process of the image recognition apparatus 100 in the present embodiment.

  As shown in FIG. 11, the recalling unit 102d controls the user to input a collation pattern, which is image data serving as a clue, via the input unit 112 or the like (step SA-41). In addition, the recalling unit 102d may cause the user to set the cooling rate γ, the initial reverse temperature β0, and the number of steps T for updating the state via the input unit 112 or the like.

  Then, the recalling unit 102d initializes the neural network, substitutes 1 for the time t, and substitutes β0 for the reverse temperature β (step SA-42).

  Then, the recall unit 102d decodes the state of the neural network and calculates a reconstruction pattern as an output (step SA-43).

  Then, the comparison unit 102e compares the input collation pattern with the output reconstruction pattern, and calculates a constraint term (step SA-44).

Then, the recalling unit 102d determines the state update probability Prob (x _i (t + 1) of each nerve cell unit according to the energy function value E (t) including the constraint term calculated by the comparison unit 102e and the inverse temperature β (t). = 1) and the neural network is updated according to the probability (step SA-45).

  Then, the recalling unit 102d determines whether or not the time t, which is the updated number of steps, has reached a predetermined number of steps T (step SA-46).

  When the predetermined number of steps T has not been reached (step SA-46, No), the recalling unit 102d increments t (t → t + 1), multiplies the inverse temperature β (t) by γ, and then β (t + 1 ) Is set (step SA-47), and the above-described steps SA-43 to SA-46 are repeated.

  On the other hand, when the predetermined number of steps T has been reached (step SA-46, Yes), the recalling unit 102d decodes the state of the neural network at that time, acquires a reconstruction pattern, and outputs it to the output unit 114. The recall process is finished (step SA-48).

[Example]
Next, an example in which the above-described embodiment is applied and image recall is performed without using mask information will be described below with reference to FIG. FIG. 12 is a diagram showing the results of image recall without using mask information according to this example.

  In this example, the constrained annealing method according to the present embodiment described above was used. The stored images used in the present embodiment are four images (Lena, Mandrill, Peppers, Tree) selected from the USC-SIPI image database of the public image database. These images are expressed in 24-bit RGB. The image size was reduced to 64 pixels in the vertical direction and 64 pixels in the horizontal direction, and then converted into binary data for use. The length of the binary data was N = 98304 bits.

  These original data were converted into a connection pattern with a block length L = 2 and stored in a self-associative memory model with a loose connection. The number of connections received by each nerve cell unit was uniformly set to K = 1500, and the connection source was randomly determined for each nerve cell unit. Further, in order to normalize the connection weight, a part of the learning rule was changed, and in the normal formula, the division by the number N ′ of nerve cell units was divided by K instead.

  The numerical calculation conditions were set such that the constraint strength was s = 0.7, the initial value of the reverse temperature was β (0) = 1.0, and the amplification factor of the reverse temperature was γ = 1.05. The initial state of the neural network was a random state. As the clue pattern, four types of complete stored data (four images from the top of the first column in FIG. 12) and two types of data with noise (two images from the bottom of the first column) were used. The noisy data is created by randomly inverting 10% of the binary data corresponding to the Lena image, and the other is a binary image obtained by filling the left 20% of the Lena image with black. Created with data. Both have the same number of inverted bits from the original data.

  When the state was updated by the constrained annealing method under the above conditions, the corresponding image output was obtained within 50 steps. Since the state update is probabilistic, the results may vary from trial to trial, but repeated trials of the constrained annealing method yielded correct images in most cases. In particular, even when the clue pattern is noisy, the corresponding Lena image can be recalled. From the above, it was shown that the corresponding storage pattern can be correctly extracted from the clue pattern by the proposed constrained annealing method without mask information. In the present embodiment, a color reconstructed image was actually obtained using a color image, but in the form of an electronic application, it is displayed in black and white in FIG. For the color image obtained as a result of this example, it is possible to refer to published papers (Makito Oku, Takaki Ei and S et al. Transactions, 2013, Volume: 24, Issue: 11, pp. 1877-1887).

  This is the end of the description of the present embodiment.

[Other embodiments]
Although the embodiments of the present invention have been described so far, the present invention is not limited to the above-described embodiments, but can be applied to various different embodiments within the scope of the technical idea described in the claims. It may be implemented.

  Moreover, although the case where the image recognition apparatus 100 performs processing in a stand-alone form has been described as an example, the image recognition apparatus 100 performs processing in response to a request from the client terminal, and returns the processing result to the client terminal. You may make it do.

  In addition, among the processes described in the embodiment, all or part of the processes described as being automatically performed can be performed manually, or the processes described as being performed manually can be performed. All or a part can be automatically performed by a known method.

  In addition, unless otherwise specified, the processing procedures, control procedures, specific names, information including registration data for each processing, parameters such as search conditions, screen examples, and database configurations shown in the above documents and drawings Can be changed arbitrarily.

  Further, regarding the image recognition apparatus 100, each illustrated component is functionally conceptual, and does not necessarily need to be physically configured as illustrated.

  For example, the processing functions provided in each device of the image recognition device 100, in particular, the processing functions performed by the control unit 102, all or any part thereof are interpreted and executed by a CPU (Central Processing Unit) and the CPU. It may be realized by a program to be executed, or may be realized as hardware by wired logic. The program is recorded on a recording medium to be described later, and is mechanically read by the image recognition apparatus 100 as necessary. That is, in the storage unit 106 such as a ROM or an HD, a computer program for performing various processes by giving instructions to the CPU in cooperation with an OS (Operating System) is recorded. This computer program is executed by being loaded into the RAM, and constitutes a control unit in cooperation with the CPU.

  The computer program may be stored in an application program server connected to the image recognition apparatus 100 via an arbitrary network 300, and may be downloaded in whole or in part as necessary. It is.

  In addition, the program according to the present invention may be stored in a computer-readable recording medium, and may be configured as a program product. Here, the “recording medium” is any memory card, USB memory, SD card, flexible disk, magneto-optical disk, ROM, EPROM, EEPROM, CD-ROM, MO, DVD, Blu-ray Disc, etc. Of “portable physical media”.

  The “program” is a data processing method described in an arbitrary language or description method, and may be in any format such as source code or binary code. The “program” is not necessarily limited to a single configuration, but is distributed in the form of a plurality of modules and libraries, or in cooperation with a separate program represented by an OS (Operating System). Including those that achieve the function. Note that a well-known configuration and procedure can be used for a specific configuration for reading a recording medium, a reading procedure, an installation procedure after reading, and the like in each device described in the embodiment.

  Various databases and the like (neuron file 106b and the like) stored in the storage unit 106 are storage means such as a memory device such as a RAM and a ROM, a fixed disk device such as a hard disk, a flexible disk, and an optical disk. Stores various programs, tables, databases, web page files, etc. used for providing websites.

  The image recognition apparatus 100 may be configured as an information processing apparatus such as a known personal computer or workstation, or may be configured by connecting any peripheral device to the information processing apparatus. The image recognition apparatus 100 may be realized by installing software (including programs, data, and the like) that causes the information processing apparatus to realize the method of the present invention.

  Furthermore, the specific form of distribution / integration of the devices is not limited to that shown in the figure, and all or a part of them may be functional or physical in arbitrary units according to various additions or according to functional loads. Can be distributed and integrated. That is, the above-described embodiments may be arbitrarily combined and may be selectively implemented.

  As described above in detail, according to the present invention, in the associative memory model neural network in which the masked pattern is stored, the corresponding pattern can be extracted without using the mask pattern. An apparatus, an image recognition method, and a program can be provided, and is particularly useful in various fields such as image recognition and image processing.

DESCRIPTION OF SYMBOLS 100 Image recognition apparatus 102 Control part 102a Mask production | generation part 102b Logic operation part 102c Learning part 102d Recall part 102e Comparison part 104 Communication control interface part 106 Memory | storage part 106b Neuron file 108 Input / output control interface part 112 Input part 114 Output part 200 External system 300 network

Claims (10)

An image recognition device having at least a storage unit and a control unit,
The storage unit
An association in which a neural network consisting of neural cell units learns a connection pattern in which a mask pattern is associated with a mask pattern that is an exclusive OR or a negative exclusive OR of an original pattern and a mask pattern of multiple images. Memorize information about the memory model,
The control unit
Recall means for obtaining a reconstructed pattern of a recall result by introducing a matching pattern, which is a specified cue image, into the associative memory model, decoding a state and obtaining a reconstructed pattern and repeating state updating;
The reconstructed pattern decoded with the state update by the recalling unit is compared with the collation pattern, and the nerve corresponding to the portion where the value is identical between the reconstructed pattern and the collation pattern For the cell unit, it is constrained to easily take the same value as the state before the update at the time of the state update. On the other hand, the state before the update at the time of the state update for the nerve cell unit corresponding to the portion where the value does not match A comparison means for constraining to make it easy to take a value different from
An image recognition apparatus comprising: The image recognition apparatus according to claim 1,
The above recall means
By annealing, the state of the neural network is converged by gradually lowering the temperature along with the state update, and the energy function value is controlled so as not to fall into a local minimum.
An image recognition apparatus. The image recognition apparatus according to claim 2,
The above recall means
Each time the state is updated, the temperature is decreased exponentially by multiplying the reverse temperature by the rate of increase.
An image recognition apparatus. The image recognition apparatus according to claim 3.
The comparison means is
In order to add the constraint, setting a constraint term that reduces the value of the energy function of the neural network in accordance with the sum of products of values between the reconstruction pattern and the matching pattern;
An image recognition apparatus. The image recognition apparatus according to claim 4.
The internal potential of the neuron unit is
The energy function with the constraint term is partially differentiated in the state of the nerve cell unit and the sign is inverted.
The state update probability of the above neuron unit is
The exponent obtained by reversing the sign of the product of the internal potential and the inverse temperature and adding 1 to the base is represented as a fraction with 1 as the numerator,
The above recall means
Recalculating the state update probability for each state update,
An image recognition apparatus. The image recognition apparatus according to any one of claims 1 to 5,
The above recall means
Setting the neural network to a random value in the initial state and updating the state of the neural network according to a synchronous and probabilistic update rule;
An image recognition apparatus. The image recognition device according to claim 1,
The mask pattern is
Be a random binary pattern,
An image recognition apparatus. The image recognition apparatus according to claim 7.
The mask pattern is
The pattern created by the mask pattern common to any number of bits in the original pattern among the plurality of images,
An image recognition apparatus. An image recognition method executed in a computer including at least a storage unit and a control unit,
The storage unit
An association in which a neural network consisting of neural cell units learns a connection pattern in which a mask pattern is associated with a mask pattern that is an exclusive OR or a negative exclusive OR of an original pattern and a mask pattern of multiple images. Remembers information about the memory model,
Executed in the control unit,
A recall process for obtaining a reconstructed pattern of a recall result by introducing a matching pattern that is a designated cue image into the associative memory model, repeating the state update while decoding the state and obtaining the reconstructed pattern, and
The reconstructed pattern decoded with the state update by the recalling unit is compared with the collation pattern, and the nerve corresponding to the portion where the value is identical between the reconstructed pattern and the collation pattern For the cell unit, it is constrained to easily take the same value as the state before the update at the time of the state update. On the other hand, the state before the update at the time of the state update for the nerve cell unit corresponding to the portion where the value does not match. A comparison procedure that adds constraints to make it easier to take different values.
An image recognition method comprising: A program for causing a computer including at least a storage unit and a control unit to be executed,
The storage unit
An association in which a neural network consisting of neural cell units learns a connection pattern in which a mask pattern is associated with a mask pattern that is an exclusive OR or a negative exclusive OR of an original pattern and a mask pattern of multiple images. Remembers information about the memory model,
Executed in the control unit,
A recall process for obtaining a reconstructed pattern of a recall result by introducing a matching pattern that is a designated cue image into the associative memory model, repeating the state update while decoding the state and obtaining the reconstructed pattern, and
The reconstructed pattern decoded with the state update by the recalling unit is compared with the collation pattern, and the nerve corresponding to the portion where the value is identical between the reconstructed pattern and the collation pattern For the cell unit, it is constrained to easily take the same value as the state before the update at the time of the state update. On the other hand, the state before the update at the time of the state update for the nerve cell unit corresponding to the portion where the value does not match. A comparison procedure that adds constraints to make it easier to take different values.
A program for running