KR20000026152A - High speed weighted fuzzy decision neural network device of quantized fuzzy trigonometric function - Google Patents

High speed weighted fuzzy decision neural network device of quantized fuzzy trigonometric function Download PDF

Info

Publication number
KR20000026152A
KR20000026152A KR1019980043551A KR19980043551A KR20000026152A KR 20000026152 A KR20000026152 A KR 20000026152A KR 1019980043551 A KR1019980043551 A KR 1019980043551A KR 19980043551 A KR19980043551 A KR 19980043551A KR 20000026152 A KR20000026152 A KR 20000026152A
Authority
KR
South Korea
Prior art keywords
fuzzy
value
block
weighted
images
Prior art date
Application number
KR1019980043551A
Other languages
Korean (ko)
Other versions
KR100298942B1 (en
Inventor
문성룡
Original Assignee
문성룡
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 문성룡 filed Critical 문성룡
Priority to KR1019980043551A priority Critical patent/KR100298942B1/en
Publication of KR20000026152A publication Critical patent/KR20000026152A/en
Application granted granted Critical
Publication of KR100298942B1 publication Critical patent/KR100298942B1/en

Links

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N5/00—Details of television systems
    • H04N5/14—Picture signal circuitry for video frequency region
    • H04N5/21—Circuitry for suppressing or minimising disturbance, e.g. moiré or halo
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/043—Architecture, e.g. interconnection topology based on fuzzy logic, fuzzy membership or fuzzy inference, e.g. adaptive neuro-fuzzy inference systems [ANFIS]

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Software Systems (AREA)
  • Biophysics (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Fuzzy Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Computational Mathematics (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Automation & Control Theory (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Compression Or Coding Systems Of Tv Signals (AREA)
  • Picture Signal Circuits (AREA)
  • Facsimile Image Signal Circuits (AREA)

Abstract

PURPOSE: A high speed weighted fuzzy decis ion neural network device is to selectively apply a weighted value and improve a calculating speed, thereby improving the quality of an image. CONSTITUTION: A high speed weighted fuzzy decis ion neural network device comprises a resister part for receiving and outputting information, a fuzzy block for receiving a resister value and expressing data into 0 and 1, non-fuzzy block for generating a 8 bit pixel value using a fuzzy average method, and a decis ion block which compares the pixel value generated from the non-fuzzy block with an estimator value and obtains a non-fuzzy value having a minimum error value and then corrects the non-fuzzy value within a limit of error obtained by comparing with the estimator value in DWW(dynamic weighted warping) and outputs the corrected value. In the fuzzy block, a circuit for eliminating a noise of image is quantized using a trigonometric function, thereby allowing the circuit to be digitally treated.

Description

양자화 퍼지 삼각함수의 고속 가중 퍼지결정 신경망 장치Fast Weighted Fuzzy Crystal Neural Network Devices of Quantization Fuzzy Trigonometric Functions

기존의 WFM(Weighted Fuzzy Mean)필터는 아날로그 방식으로 퍼지 함수를 적용하였고, 속도가 느린 단점을 가지고 있다Existing WFM (Weighted Fuzzy Mean) filter has applied fuzzy function by analog method and has a disadvantage of slow speed.

본원 발명은 우선 데이터의 디지털 처리를 위한 퍼지 함수의 양자화를 제안하고, 가중 FDNN(Fuzzy Decision Neural Network)구조를 갖도록 설계하였고, 양자화 오차를 최소화하기 위하여 DWW(Dynamic Weighted Warping)알고리즘을 적용 자율학습 방법의 회로를 설계 집적화함으로써 영상의 화질을 개선하기 위한 것이다.The present invention proposes a quantization of a fuzzy function for digital processing of data, designed to have a weighted FDNN (Fuzzy Decision Neural Network) structure, and applies a DWW (Dynamic Weighted Warping) algorithm to minimize quantization error. It is to improve the image quality of the image by designing and integrating the circuit.

도 1은 전체 회로의 하드웨어 블록도1 is a hardware block diagram of an entire circuit

도 2는 비퍼지화 블록도2 is an unfuzzy block diagram

도 3은 결정 블록의 블록도3 is a block diagram of a decision block

도 4는 DWW 블록도4 is a DWW block diagram

본원 발명은 회로를 구성함에 있어서 특수한 기능을 수행하는 블록, 레지스터 블록, 퍼지화 블록, 비퍼지화 블록, 평가자 블록, 결정 블록으로 구성되어 있다. 각 블록은 합성되어 영상에서의 고밀도 잡음에서 양호한 특성을 나타내면서 잡음로를 제거하게 된다. 각 블록은 해당하는 연산을 수행하고 이를 다음 블록으로 전송하는 방법으로 구성되어 양자화 오차를 DWW알고리즘을 이용하여 효과적으로 감소할 수 있다. 그리고 FDNN구조를 갖는 WFM필터의 성능은 아날로그구현한 것보다 우수하다.The present invention consists of a block, a register block, a fuzzy block, an unfuzzy block, an evaluator block, and a decision block that perform special functions in constructing a circuit. Each block is synthesized to remove noise paths while exhibiting good characteristics in high-density noise in the image. Each block is configured by performing a corresponding operation and transmitting it to the next block so that the quantization error can be effectively reduced by using a DWW algorithm. And the performance of WFM filter with FDNN structure is better than analog implementation.

이와 같은 본원 발명을 상세히 설명하면 다음과 같다.The present invention will be described in detail as follows.

레지스터 블록은 픽셀 정보를 입력받고 이를 레지스터로 출력한다. 그 과정은 쉬프트를 이용하게 된다. 인터페이스를 위해서 칩 enable단자를 제어 단자로 사용하였다.The register block receives pixel information and outputs it to the register. The process uses shifts. The chip enable terminal is used as the control terminal for the interface.

퍼지화 블록에서 퍼지 함수라 함은 데이터를 0과 1사이의 값으로 나타내는 함수를 퍼지 함수라고 하는데, 그 함수에는 삼각 퍼지 함수, 사다리꼴 퍼지 함수, L-R형태의 퍼지 함수 등이 있다. 이 함수의 결과를 그대로 회로로 옮기기에 디지털은 계산에 어려움이 많다. 따라서 삼각 퍼지 함수를 사용하고, 양자화하는 방법을 제안하고 적용한 결과를 도 2에 표시하였다.In the fuzzy block, a fuzzy function is a function that represents data between 0 and 1, called a fuzzy function, which includes a triangular fuzzy function, a trapezoidal fuzzy function, and an L-R type fuzzy function. Since the result of this function is transferred to the circuit as it is, digital is difficult to calculate. Therefore, the results of the proposed and applied method using a triangular fuzzy function and quantization are shown in FIG.

제안한 삼각 퍼지 함수의 구간은 DARK, MIDDLE, BRIGHT이다. 퍼지화 블록의 설계는 양자화 구간에 따라 매핑하는 방법으로 퍼지값을 산출하였다. 양자화함으로써 회로의 디지털 처리가 가능해졌다.The proposed triangular fuzzy intervals are DARK, MIDDLE, and BRIGHT. In the design of the fuzzy block, a fuzzy value was calculated by mapping according to the quantization interval. Quantization allows digital processing of the circuit.

비퍼지화 블록은 다음과 같다. 결과의 비퍼지 값은 퍼지 평균법을 이용하여 구하였고 그 결과 픽셀 값을 갖게 된다. 이 값은 결정 블록으로 전송되어 평가자와 비교되어 최소값을 갖는 값을 선택하게 된다.The non-fuzzy block is as follows. The non-fuzzy values of the result were obtained using the fuzzy averaging method, resulting in pixel values. This value is sent to the decision block and compared with the evaluator to select the value with the minimum value.

비퍼지 블록의 퍼지 규칙과 블록도는 그림 3과 같다. 비퍼지 값은 레지스터 출력중 픽셀 값을 취하여 계산하여 그 결과를 출력한다.The fuzzy rule and block diagram of the non-fuzzy block are shown in Figure 3. The non-fuzzy value is calculated by taking the pixel value during register output and outputting the result.

결정 블록을 상세히 설명하면 다음과 같다.The decision block is described in detail as follows.

비퍼지 출력값은 결정 블록으로 전송되어 평가자 값과 비교하게 된다. 평가자 값은 레지스터 출력의 픽셀 값을 퍼지 평균법으로 구한 평가자 값과 비교하게 된다. 비퍼지 결과값은 평가자 값과 절대값 연산을 하고, 식별값을 더한다. 이 식별값은 디코더를 통해서 어느 값이 최소값을 갖는지를 표시하게 된다. 다음으로 최소값을 찾게 되면 평가자 값과 오차가 가장 적은 값을 얻는 결과가 된다. 이 값의 식별값으로 원래 데이터를 디코딩하게 되면 최소값을 갖는 픽셀 데이터를 찾을 수 있다. 이 값은 오차 이내로 될 때 까지 DWW블록에서 조정된다.The unfuzzy output is sent to a decision block to compare with the evaluator value. The evaluator value is compared with the evaluator value obtained by the fuzzy averaging method. The non-fuzzy results are computed with the evaluator's value and the absolute value, and the identified value is added. This identification will indicate through the decoder which value has the minimum value. Next, finding the minimum value results in obtaining the value of the evaluator and the least error. Decoding the original data with the identification of this value will find the pixel data with the minimum value. This value is adjusted in the DWW block until it is within error.

DWW알고리즘에 대해 상세히 설명하면 다음과 같다. DWW알고리즘은 임의의 입력영상과 일정한 영상이 저장된 기준영상을 정규화한 후 이들 영상을 분할하여 두 영상사이에 있는 오차를 일련의 조건으로 제한하는 비선형 warping함수에 의해 영상의 가중치를 비교하여 일치하도록 하는 알고리즘이다. 이 warping함수는 가중치 허용오차의 값을 최소로 하는 경로에 따라 유동성 있게 결정할 수 있다.The DWW algorithm is described in detail as follows. The DWW algorithm normalizes a reference image that stores arbitrary input images and constant images, divides them, and compares the weights of the images by a nonlinear warping function that limits the error between the two images to a set of conditions. Algorithm. This warping function can be determined flexibly along the path that minimizes the value of the weight tolerance.

만일 두 영상사이에 오차가 없다면 warping함수는 대각선과 일치하며, 오차가 커지면 이 함수의 점들은 대각선으로부터 벗어나게 된다. 두 벡터 성분의 값의 차이는 이들간의 거리에 의해 얻어질 수 있다.If there is no error between the two images, the warping function coincides with the diagonal, and if the error increases, the points of this function deviate from the diagonal. The difference in the values of the two vector components can be obtained by the distance between them.

이와 같은 방법에 의한 본 발명은 선택적으로 가중치를 적용하여 계산속도를 향상시킴으로써 영상에서의 잡음을 효과적으로 제거하고, 영상의 화질을 향상시킴과 동시에 동영상의 전송상의 잡음도 또한 제거할 수 있다.The present invention by such a method can selectively remove the noise in the image by improving the computational speed by selectively applying the weight, improve the image quality and at the same time can also remove the noise in the transmission of the video.

Claims (3)

영상의 잡음을 제거하는 회로에 있어서,In a circuit for removing noise in an image, 정보를 입력받고 쉬프트를 이용하여 출력하는 레지스터부와;A register unit for receiving information and outputting the shift; 상기 레지스터의 값을 받아 데이터를 0과 1의 값으로 나타내는 퍼지화블록과; 퍼지평균법을 이용하여 8비트의 픽셀 값을 발생시키는 비퍼지화블록과; 상기 비퍼지화블록에서 발생된 픽셀값과 평가자와 비교하여 최소 오차 값을 갖는 비퍼지값을 MIN블록에서 찾고 DWW블록에서 평가자 값과 비교하여 선정한 오차 이내로 수정되어 최종 픽셀정보를 출력하는 결정블록으로 구성된 것을 특징으로하는 양자화 퍼지 삼각 함수의 고속 가중 FDNN장치.A fuzzy block that receives a value of the register and represents data as values of 0 and 1; A non-fuzzy block for generating an 8-bit pixel value using the fuzzy averaging method; As a decision block that outputs the final pixel information by finding a non-fuzzy value having a minimum error value in the MIN block and comparing it with the evaluator value in the DWW block by comparing the pixel value generated in the non-fuzzy block with the evaluator. A fast weighted FDNN device of quantization fuzzy trigonometric function, characterized in that configured. 청구 1항의 레지스터의 값을 받아 데이터를 0과 1의 값으로 나타내는 퍼지화블록에 있어서,In the fuzzy block which receives the value of the register of claim 1 and represents data as a value of 0 and 1, 삼각 퍼지 함수를 사용하여 양자화 함으로써 회로의 디지털 처리가 가능하게 하는 것을 특징으로하는 양자화 퍼지 삼각 함수의 고속 가중 FDNN 장치.A fast weighted FDNN device of quantization fuzzy trigonometric functions, characterized in that digital processing of a circuit is possible by quantizing using a triangular fuzzy function. 임의의 입력영상과 일정한 영상이 저장된 기준영상을 정규화한 후 이들 영상을 분할하여 두 영상사이에 있는 오차를 일련의 조건으로 제한하는 비선형 warping 함수에 의해 영상의 가중치를 비교하여 일치하도록 DWW 알고리즘을 사용한 것을 특징으로 하는 양자화 퍼지 삼각 함수의 고속 가중 FDNN장치.Using the DWW algorithm to compare and match the weights of images by a nonlinear warping function that normalizes the reference images that store arbitrary input images and constant images, and divides them to limit the error between the two images as a series of conditions. A fast weighted FDNN device of quantization fuzzy trigonometric function, characterized in that.
KR1019980043551A 1998-10-19 1998-10-19 Fuzzy Crystal Neural Network Implementation of Quantization Fuzzy Trigonometric Functions KR100298942B1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
KR1019980043551A KR100298942B1 (en) 1998-10-19 1998-10-19 Fuzzy Crystal Neural Network Implementation of Quantization Fuzzy Trigonometric Functions

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
KR1019980043551A KR100298942B1 (en) 1998-10-19 1998-10-19 Fuzzy Crystal Neural Network Implementation of Quantization Fuzzy Trigonometric Functions

Publications (2)

Publication Number Publication Date
KR20000026152A true KR20000026152A (en) 2000-05-15
KR100298942B1 KR100298942B1 (en) 2001-10-27

Family

ID=19554425

Family Applications (1)

Application Number Title Priority Date Filing Date
KR1019980043551A KR100298942B1 (en) 1998-10-19 1998-10-19 Fuzzy Crystal Neural Network Implementation of Quantization Fuzzy Trigonometric Functions

Country Status (1)

Country Link
KR (1) KR100298942B1 (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20180129211A (en) * 2017-05-25 2018-12-05 삼성전자주식회사 Method and apparatus for quantizing data in a neural network
KR101982941B1 (en) * 2017-12-18 2019-08-28 연세대학교 원주산학협력단 Method and Apparatus for removing artifact in CT image using fuzzy neural network

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112218094A (en) * 2019-07-11 2021-01-12 四川大学 JPEG image decompression effect removing method based on DCT coefficient prediction

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20180129211A (en) * 2017-05-25 2018-12-05 삼성전자주식회사 Method and apparatus for quantizing data in a neural network
KR101982941B1 (en) * 2017-12-18 2019-08-28 연세대학교 원주산학협력단 Method and Apparatus for removing artifact in CT image using fuzzy neural network

Also Published As

Publication number Publication date
KR100298942B1 (en) 2001-10-27

Similar Documents

Publication Publication Date Title
KR0176601B1 (en) Picture quality improving method & circuit using low-filtering and histogram equalization
JPH0432587B2 (en)
JPS62216476A (en) Picture processor
KR100206319B1 (en) Local contrast improving method and apparatus of a video signal
KR970000767B1 (en) Blind equalizer
JP3103914B2 (en) Data rounding circuit and data restoration circuit
US5241387A (en) Noise-reducing filter apparatus for decoded digital video signal
KR100555866B1 (en) Apparatus and method for smoothing signal by pattern adaptive filtering
KR20000026152A (en) High speed weighted fuzzy decision neural network device of quantized fuzzy trigonometric function
JP3149729B2 (en) Block distortion removal device
EP0786741B1 (en) Method and apparatus for binary coding of image data
US5612746A (en) Block matching for picture motion estimation using offset quantized pixels
JP2841362B2 (en) High efficiency coding device
JPH04297962A (en) Method and apparatus for emphasizing image
JP3791029B2 (en) Image signal processing apparatus and method
JP3020971B2 (en) Image processing device
JP2683181B2 (en) Color image processing equipment
KR100712376B1 (en) Data processing apparatus and data processing method
JP2716618B2 (en) Image coding method
JP3769770B2 (en) Quantizer and quantization method
KR100335614B1 (en) Method and apparatus for spreading error in color reproducing system
US5815207A (en) Differential coding and decoding method and related circuits
JPH09107472A (en) Method and device for image data conversion
KR0150164B1 (en) Quantization method and apparatus using error diffusion for image processing system
JPH03274863A (en) Image processing method

Legal Events

Date Code Title Description
A201 Request for examination
E902 Notification of reason for refusal
E701 Decision to grant or registration of patent right
GRNT Written decision to grant
FPAY Annual fee payment

Payment date: 20040601

Year of fee payment: 4

LAPS Lapse due to unpaid annual fee