TWI454940B - Method for designining fuzzy membership functions of of fuzzy controller auto focus - Google Patents

Description

Design method of fuzzy attribution function for fuzzy controller of autofocus

The present invention relates to an autofocus method, and more particularly to a home function design method for a fuzzy controller for autofocus.

Autofocus technology is widely used in the field of digital cameras, and the speed of focusing is a big factor in the reasons consumers consider buying a camera. The conventional autofocus step is to take the digital image captured, calculate the sharpness through the computer, and then use the autofocus controller to change the focal length or object distance to present the clearest image.

Please refer to FIG. 1 , which is a schematic diagram of a lens focusing method according to a conventional technique. The following describes the current auto-focusing search method. The global search method uses the minimum step size to perform the global-wide sharpness value search. After all the search is completed, the lens returns to the point where the sharpness value is the largest to complete the focus. The binary search method is to start searching with a large step. Once the search for the sharpness value is less than the previous time, the search is repeated with half of the step search, so that the search for the focus is repeated. The rule-based search rule is to take the ratio of the difference value of the two points to the fixed moving distance. When the difference is larger, the moving step is larger, otherwise the step is reduced, and finally, the point where the sharpness value is maximum is returned.

However, the above methods all belong to the method of quantitative movement. Because of the excessive amount of images captured, the calculation amount is huge, or the lens must be moved back and forth, so as to narrow the search pitch, and the total moving distance is too large, which is time consuming, so the auto focus search technology There is still room for improvement.

Therefore, it has been proposed to use the fuzzy theory of intelligent control theory to design the controller of the autofocus system to improve the above-mentioned problem of too many focus times. Taiwan Patent Publication No. 200712594 discloses a fuzzing device capable of blurring an input signal, and then inputting the fuzzified value with a push On the rule inference engine library, then the inference value is defuzzified to the corresponding output device, and the motor that will make the lens move is controlled.

Please refer to Figure 2, which is a block diagram of the architecture of the existing fuzzy controller. The reference numeral 100 in FIG. 2 represents a conventional fuzzy controller. The fuzzy controller 100 includes a fuzzifier 120, a fuzzy rule base 140, and a fuzzy inference engine (Fuzzy Inference Engine). 160, and Defuzzifier 180. The fuzzification mechanism 120 is configured to convert external input data (such as the aforementioned sharpness value) into appropriate fuzzification information, which is usually converted into a fuzzy calculation by using a membership function. The amount of blur, such as a language item. The fuzzy inference engine 160 is the core of the fuzzy controller 100, which solves the problem according to the obtained fuzzy information and pre-stores the fuzzy rules needed to solve the problem in the fuzzy rule base 140, and simulates the way human thinking decisions are made. Finally, the defuzzification mechanism 180 converts the fuzzy information inferred by the fuzzy inference engine 160 into clear information that is acceptable to the outside world, such as the voltage or moving position required for the lens motor to move.

However, the fuzzy controller's attribution function is designed with Try and Error, which often takes a lot of time, and the resulting attribution function can not be determined to achieve the best results, often making the auto focus It takes too long.

In view of this, it is necessary to improve the prior art to overcome the shortcomings of the prior art.

The object of the present invention is to provide a method for designing a fuzzy attribution function, which utilizes the Taguchi method and a genetic algorithm to optimize the attribution function to obtain the best fuzzy controller of the autofocus system. Combination of attribution functions to solve the above problem question.

To achieve the above object, the present invention provides a method for designing a fuzzy attribution function for autofocus, which is used to design a home function of a fuzzy controller for autofocus, which includes the following steps: (A) selecting a plurality of first language items, a plurality of second language items, and a plurality of third language items; (B) selecting a plurality of levels corresponding to the second language items and the respective third language items (C) establish one of the level values of the Taguchi orthogonal table; (D) conduct an experiment according to the Taguchi orthogonal table to obtain a plurality of experimental results, and calculate respective signal to noise ratios of the respective experimental results; E) calculating, according to the signal noise ratios, a response table corresponding to one of the second language items and the third language items; and (F) selecting, according to the response table, the respective corresponding corresponding to the second language items. One of the level values, and one of the level values corresponding to the third language items, respectively, is used as a plurality of parameters of the attribution function.

In step A, a fuzzy rule base is established, and in step A, a fuzzy rule base is further included to establish a rule corresponding to the input to the output. Specifically, a rule for mapping the combination of the first language items and the second language items to the third language items is established. Specifically, the first language items correspond to a sharpness value, and the second language items correspond to a sharpness value change amount, and the third language items correspond to a lens focal length change amount.

In a preferred embodiment, the first language items are each a value of approximately zero, a small and a positive, and the second and third language terms are each negative, negative, approximately zero, positive, and Zhengda and so on. Accordingly, the Taguchi straight line table is an L ₁₈ (2 ¹ × 3 ⁷ ) orthogonal table, and the results of these experiments are the average number of focusing times. In addition, the calculation of the respective signal murmur ratios of the results of the respective experiments is performed according to the formula of the small characteristic. It should be noted that in step F, the level value corresponding to the larger value in the response table is selected as the parameter of the attribution function.

In addition to this, step F further comprises a step G of performing a genetic algorithm on the parameters, including replication, mating and mutation steps.

According to the design method of the fuzzy attribution function for autofocus according to the present invention, the disadvantage that the attribution function obtained by the trial and error method cannot be optimized can be solved. Experiments were conducted at these level values through the Taguchi Direct Meter to obtain individual signal to noise ratios. Finally, through the establishment of the response table, and according to the larger value in the response table, the corresponding level value is used as the parameter of the attribution function. In addition, these parameters are more accurately calculated by the genetic algorithm to optimize the parameters of the attribution function. The autofocus controller designed by this method can achieve fast capture of clear images.

In order to make the above description of the present invention more comprehensible, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings.

A preferred embodiment of the method for designing the fuzzy attribution function of the present invention will be described in detail below with reference to the accompanying drawings. The method is preferably used to design the attribution function of the fuzzy controller of the autofocus. However, the present invention is not limited thereto, and the method can also be used to design other fuzzy functions of the fuzzy theory, which has a attribution function. Referring to FIG. 3, FIG. 3 is a flow chart showing a method for designing a fuzzy attribution function according to a preferred embodiment of the present invention; the method begins in step S10.

Referring to FIG. 4, FIG. 4 illustrates a definition value of a sharpness value, a sharpness value change, and a lens focal length change amount. In the preferred embodiment, the definition value, the change amount of the sharpness value, and the attribution function of the focal length change of the lens are defined as a belonging function MTF 10, a belonging function dMTF 20, and a attribution function Output, respectively. 30, and the degree of attribution is 1. In addition, two input variables are defined as the sharpness value and the sharpness value change amount, and the output is defined as the lens focal length change amount.

In step S10, a plurality of first language items, a plurality of second language items, and a plurality of third language items are selected. Specifically, the first language items correspond to the sharpness value, the second language items correspond to the sharpness value change amount, and the third language items correspond to the lens focal length change amount. In short, the sharpness value and the sharpness value change amount are blurred by the blurring means 120. Since the resolution value has no negative value, the first language terms corresponding to the selected sharpness value are each approximately zero (Approximately Zero, ZR), Positive small (PS), and Positive Big (PB). , where ZR stands for the least, PS stands for a little clarity, and PB stands for it. The second language terms corresponding to the selected change in the value of the sharpness value are negative Big (NB), Negative Small (NS), approximate zero ZR, positive small PS, and Zhengda PB, wherein ZR represents a change. The minimum amount, NB represents the larger negative value, NS represents the smaller negative value, PB represents the larger positive value, and PS represents the smaller positive value. The third language terms corresponding to the selected focal length change of the lens are negative NB, negative small NS, approximate zero ZR, positive small PS and positive large PB, wherein ZR represents the minimum step size to be moved, and NB represents The steps to be moved are the large step in the opposite direction, the step in which NS represents the moving step is the small step in the opposite direction, the step in which PB represents the moving step is the large step in the positive direction, and the step in which PS represents the moving direction is the positive direction. Small step.

Please refer to FIG. 5, which is a fuzzy rule base according to a preferred embodiment of the present invention. In step S10, a fuzzy rule base is further included, which is used to establish a plurality of rules corresponding to the input to the output, and the input generates an output by using the rules (ie, the input is subjected to the fuzzy inference engine) The resulting fuzzy language item). Here, a rule for mapping the input values of the first language items combined with the second language items to the third language items is established. In the preferred embodiment, the fuzzy rule base represents a rule for outputting two input variables (the attribution function MTF 10 and the attribution function dMTF 20) to the attribution function Output 30, wherein the subscripts only represent the labels of the rules. For example, when the first language item of the input attribution function MTF 10 is PB and the second language item of the attribution function dMTF 20 is NS, the third language item of the output attribution function Output 30 is the fuzzy rule base. Rule 2, ie ZR ₂ . In vernacular, when the sharpness value is very clear (PB) and the change in the sharpness value is not large (NS), the next shot change should be close to zero (ZR), ie it is very close focus.

In step S20, a plurality of level values corresponding to the second language items and the third language items are selected. Specifically, NB, NS, PS, and PB of the selected sharpness value change amount dMTF 20 and the lens focal length change amount Output 30 are control factors, and the rest are fixed values. In order to clearly indicate that they are each represented by control factors A to H. As shown in FIG. 4, in the preferred embodiment, the parameter values of the ZR, PS, and PB of the first language item are fixed to 0.3, 0.6, and 0.9, and the ZR of the second language item and the third language item are The ZR parameter is fixed to zero. However, the invention is not limited to this selected mode. Figure 6 is a plurality of level values selected in accordance with a preferred embodiment of the present invention. Please refer to Figure 6. For example, the level values of the control factors are selected as shown in FIG. 6.

In step S30, one of the level values is established as an Orthogonal array. Figure 7 is a diagram of the Taguchi orthogonal table of the level values of the preferred embodiment of the present invention. Please refer to FIG. 7. In the preferred embodiment, the field line orthogonal table adopts an L ₁₈ (2 ¹ × 3 ⁷ ) orthogonal table, wherein the subscript 18 represents 18 sets of experiments, and the superscript indicates the number of control factors ( This is 8 of the NB, NS, PS and PB of the second language item and the third language item, and the number represents the number of the level of the control factor.

In step S40, an experiment is performed according to the Taguchi orthogonal table to obtain a plurality of experimental results, And calculate the respective Signal Noise Ratio (S/N Ratio) of each of the experimental results. In the preferred embodiment, the experimental results are the average number of focusing times, that is, the attribution function MTF 10 selected by the parameters of the field intersection table, the attribution function dMTF 20, and the attribution function Output 30 are utilized. This fuzzy controller performs a focus experiment and averages the number of times the focus is completed. In addition, the respective S/N ratios of the respective experimental results are calculated according to the Smaller the Better characteristic formula.

In step S50, a response table corresponding to the second language items and the third language items is calculated according to the signal noise ratios. Specifically, this is to use the characteristics of the S/N ratio to obtain the average optimal value and the minimum variation of the quality characteristics. The larger the S/N ratio, the smaller the loss, the level combination and quality of the parameter. The characteristics are optimal. Please refer to Fig. 8. Fig. 8 is a response table calculated according to the S/N ratio of Fig. 7.

In step S60, selecting one of the level values corresponding to each of the second language items according to the response table, and one of the level values corresponding to each of the third language items, A plurality of parameters as a attribution function. Specifically, as shown in FIG. 8, the level value 1 of the control factor A is larger than the level value 2, so the control factor A is selected as the level value 1, that is, the NB of the attribution function dMTF 20 is -0.18. The level value 1 of the control factor B is greater than the level values 2 and 3, so the control factor B is selected as the level value 1, i.e., the NS of the attribution function dMTF 20 is -0.1. Similarly, the PS and PB of the attribution function dMTF 20 are 0.08 and 0.23, respectively. Similarly, the NB, NS, PS, and PB of the attribution function Output 30 are -20, -15, 20, and 40, respectively.

Accordingly, the attribution function designed by the preferred embodiment of the present invention will be as shown in FIG. 9, and FIG. 9 illustrates the attribution function designed by the design method of the preferred embodiment of the present invention. That is, the attribution letter is redesigned according to the level values of the selected second language item and the third language item. The number dMTF 20 and the attribution function Output 30, and through the fuzzy controller having this attribution function, can achieve the purpose of fast focusing.

It should be noted that the level values of the second language item and the third language item designed by the present invention can also be processed by a genetic algorithm familiar to those skilled in the art, wherein the genetic algorithm includes the copying, Mating and mutation steps are not described here. Through this gene algorithm, these parameters can be further fine-tuned to achieve the optimal design of the attribution function.

In addition, the present invention also discloses an autofocus fuzzy controller, which includes a attribution function designed by the above design method, so that the performance of the autofocus controller of the fuzzy controller using the attribution functions can reach the maximum The good effect, in turn, minimizes the number of times the lens is moved, thereby overcoming the shortcomings of the autofocus time caused by the attribution function generated by the trial and error method.

In summary, the design method of the fuzzy attribution function of the present invention can solve the disadvantage that the attribution function obtained by the trial and error method cannot be optimized. Experiments were conducted with the selected level values through the Taguchi Direct Meter to obtain individual signal-to-noise ratios that can be used as quality characteristics of the measurement system. The S/N ratio is the best, which is the best combination of parameters. The products produced under this parameter level have the least loss and the smallest variability. Finally, through the establishment of the response table, the level value corresponding to the larger value in the response table is used as the parameter of the attribution function. In addition, these parameters are more accurately calculated by the genetic algorithm to optimize the parameters of the attribution function. The autofocus controller designed by this method will achieve the purpose of quickly capturing clear images.

While the invention has been described above in terms of the preferred embodiments, the invention is not intended to limit the invention, and the invention may be practiced without departing from the spirit and scope of the invention. Retouching, so the scope of protection of the present invention is attached to the patent application The definition of the enclosure shall prevail.

10‧‧‧ attribution function MTF

20‧‧‧ attribution function dMTF

30‧‧‧ attribution function Output

100‧‧‧Fuzzy controller

120‧‧‧Fuzzy institutions

140‧‧‧Fuzzy Rule Base

160‧‧‧Fuzzy inference engine

180‧‧‧Unfuzzification

S10‧‧‧ steps

S20‧‧‧ steps

S30‧‧‧ steps

S40‧‧‧ steps

S50‧‧ steps

S60‧‧ steps

S70‧‧‧ steps

FIG. 1 is a schematic diagram of a lens focusing mode of the prior art.

Figure 2 is a block diagram of the architecture of the existing fuzzy controller.

FIG. 3 is a flow chart showing a method for designing a fuzzy attribution function according to a preferred embodiment of the present invention.

Figure 4 is a graph showing the attribution function of a sharpness value, a change in the sharpness value, and a change in the focal length of a lens.

Figure 5 is a fuzzy rule base of a preferred embodiment of the present invention.

Figure 6 is a selection of a plurality of levels of values in accordance with a preferred embodiment of the present invention.

Figure 7 is a top-of-the-line intersection of the level values of the preferred embodiment of the present invention.

Figure 8 is a response table calculated from the S/N ratio of Figure 7.

Figure 9 is a diagram showing the attribution function designed by the design method of the preferred embodiment of the present invention.

S10‧‧‧ steps

S20‧‧‧ steps

S30‧‧‧ steps

S40‧‧‧ steps

S50‧‧ steps

S60‧‧ steps

S70‧‧‧ steps

Claims (8)

A method for designing a fuzzy attribution function of a fuzzy controller for autofocus, which is used for designing a attribution function of an autofocus fuzzy controller, comprising the following steps: (A) selecting a plurality of first language items, a plurality of a second language item and a plurality of third language items, wherein the first language items correspond to a sharpness value, the second language items correspond to a sharpness value change amount, and the third language items Corresponding to a lens focal length change; (B) selecting a plurality of level values corresponding to the second language items and the third language items; (C) establishing one of the level values of the Taguchi orthogonal table; (D) using the fuzzy controller to perform a focusing experiment according to the Taguchi orthogonal table to obtain a plurality of experimental results, wherein the experimental results are average focusing times, and calculating respective signal to noise ratios of the respective experimental results; (E) calculating, according to the signal noise ratios, a response table corresponding to the second language items and the third language items; and (F) selecting the second language items according to the response table. Corresponding to these level values One of the plurality of standard values, and the plurality of third language item corresponding to respective one of those, as a plurality of parameters of the membership function. The method for designing a fuzzy attribution function of a fuzzy controller for autofocus according to claim 1, wherein the first language term values are approximately zero, positive, and positive, and the second And the third language term values are negative, negative, approximate zero, positive and positive. The method for designing a fuzzy attribution function of a fuzzy controller for autofocus according to claim 1, wherein the step A includes establishing a fuzzy rule base for using the first language item and the The combination of the second language items maps to the rules of the third language items. The method for designing a fuzzy attribution function of a fuzzy controller for autofocus according to claim 2, wherein the Taguchi orthogonal table is an L ₁₈ (2 ¹ × 3 ⁷ ) orthogonal table. The method for designing a fuzzy attribution function of a fuzzy controller for autofocus according to claim 1, wherein the step of calculating a respective signal noise ratio of each of the experimental results is based on a small characteristic formula Come on. The method for designing a fuzzy attribution function for a fuzzy controller for autofocus according to claim 1, wherein in step F, a level value corresponding to a larger value in the response table is selected as a parameter of the attribution function. The method for designing a fuzzy attribution function for a fuzzy controller for autofocus according to claim 1, further comprising performing a genetic algorithm on the parameters, including a copying, mating, and a mutation step. An autofocus fuzzy controller comprising a attribution function designed by the method of claim 1 of the patent application.