KR950006643A

KR950006643A - Scale factor adjustment circuit

Info

Publication number: KR950006643A
Application number: KR1019930017538A
Authority: KR
Inventors: 최광철
Original assignee: 김광호; 삼성전자 주식회사
Priority date: 1993-08-31
Filing date: 1993-08-31
Publication date: 1995-03-21
Also published as: KR100224801B1

Abstract

본 발명은 신경회로망을 이용한 스케일 팩터 조정회로를 공개한다.The present invention discloses a scale factor adjustment circuit using neural networks.

그 회로는 2프레임의 화소 데이타를 저장하기 위한 제1저장수단, 상기 제1저장수단의 제1프레임분의 화소데이타를 순서대로 입력하여 예상되는 부호길이와 예상되는 스케일 팩터를 출력하기 위한 신경망, 상기 신경망의 출력신호를 입력하여 각 화소 데이타에 대한 예상되는 스케일 팩터와 예상되는 부호 길이를 저장하기 위한 제2, 제3저장수단, 상기 제3저장수단에 예상되는 부호길이가 저장되는 동시에 각 예상되는 부호길이를 누산하기 위한 누산수단, 상기 제2, 제3저장수단에 저장된 예상되는 부호길이 및 예상되는 스케일 팩터와 상기 누산수단의 예상되는 총 부호길이를 입력하여 정규화하여 계산된 스케일 펙터와 원하는 부호길이를 출력하는 정규화 수단; 상기 제1프레임의 화상 데이타를 이산 코사인 변환을 하고 양자화하여 출력되는 부호 길이와 상기 원하는 부호길이의 차를 계산하여 오차값을 발생하고 상기 계산된 스케일 팩터를 상기 오차값만큼 보상하여 최정적인 스케일 팩터를 발생하기 위한 필터를 구비하여 상기 최종적인 스케일 팩터에 따라 양자화를 수행하는 것을 특징으로 한다. 따라서, 보다 정확한 스케일 팩터를 발생할 수가 있다.The circuit comprises first storage means for storing pixel data of two frames, neural network for outputting an expected code length and an expected scale factor by inputting pixel data for the first frame of the first storage means in order; Second and third storage means for storing the expected scale factor and the expected code length for each pixel data by inputting the output signal of the neural network, and the expected code length is stored in the third storage means An accumulating means for accumulating the code length, an expected code length and an expected scale factor stored in the second and third storage means, and a scale factor calculated and normalized by inputting the estimated total code length of the accumulating means. Normalization means for outputting a code length; Discrete cosine transform the quantized image data of the first frame and calculate the difference between the output code length and the desired code length to generate an error value, and compensate the calculated scale factor by the error value to obtain an optimal scale factor. And a filter for generating a quantization according to the final scale factor. Thus, a more accurate scale factor can be generated.

Description

Scale factor adjustment circuit

본 내용은 요부공개 건이므로 전문내용을 수록하지 않았음Since this is an open matter, no full text was included.

제1도는 본 발명의 일실시예의 신경회로망을 이용한 화상 압축기의 스케일 팩터 조정회로를 나타내는 것이다.1 shows a scale factor adjustment circuit of an image compressor using a neural network of an embodiment of the present invention.

제2도는 본 발명의 다른 실시예의 신경회로망을 이용한 화상 압축기의 스케일 팩터 조정회로를 나타내는 것이다.2 shows a scale factor adjustment circuit of an image compressor using a neural network of another embodiment of the present invention.

Claims

First storage means for storing pixel data for two frames; A first neural network for sequentially inputting pixel data of the first frame of the storage means to output an expected code length of each pixel data; Second storage means for storing the expected code length in order; Accumulating means for accumulating the expected code length; While the first neural network outputs the expected code length for the pixel data of the second frame of the first storage means, the pixel data of the first frame of the first storage means is sequentially input to output the expected scale factor. A second neural network for; Normalization means for inputting and normalizing an expected code length and an expected total code length to the scale factor and the output signal of the first neural network to perform a calculated scale factor and a desired code length; Discrete cosine transform and quantize the image data of the first frame to generate an error value by calculating the difference between the code length and the desired code length, and compensate the calculated scale factor by the error value to obtain a final scale factor. And a filter for generating a quantization according to the final scale factor.

First storage means for storing pixel data of two frames; A neural network for inputting pixel data for the first frame of the first storage means in order to output an expected code length and an expected scale factor; Second and third storage means for inputting an output signal of the neural network to store an expected scale factor and an expected code length for each pixel data; Accumulating means for accumulating the expected code lengths while storing the expected code lengths in the third storage means; Normalization means for inputting and normalizing an expected code length and an expected scale factor stored in the second and third storage means and an expected total code length of the accumulation means to output a calculated scale factor and a desired code length; Discrete cosine transform and quantize the image data of the first frame to generate an error value by calculating the difference between the code length and the desired code length, and compensate the calculated scale factor by the error value to obtain a final scale factor. And a filter for generating a circuit to perform quantization according to the final scale factor.

※ Note: The disclosure is based on the initial application.