-
公开(公告)号：KR1020120072151A
公开(公告)日：2012-07-03
申请号：KR1020100133968
申请日：2010-12-23
Applicant: 전남대학교산학협력단
IPC: G10L17/00
Abstract: PURPOSE: A method for calculating reliability based on a log-likelihood score is provided to obtain the best reliability value without predicting a too big or small value which is reflected to a voice and an image. CONSTITUTION: Each voice score and each image score are normalized in the range of 0 to 1(S1200). The difference between a first voice score and a second voice score is calculated by voice reliability(S1300). The first voice score is the maximum value among the normalized voice scores. The difference between a first image score and a second image score is calculated by image reliability. The maximum value of the second voice score is limited by a critical voice score. The maximum value of the second image score is limited by a critical image score.
Abstract translation: 目的：提供一种基于对数似然分数计算可靠性的方法，以获得最佳的可靠性值，而无需预测反映到语音和图像的太大或较小的值。 构成：每个语音分数和每个图像分数在0到1的范围内归一化（S1200）。 通过语音可靠性计算第一语音分数和第二语音分数之间的差异（S1300）。 第一语音分数是归一化语音分数中的最大值。 通过图像可靠性计算第一图像分数和第二图像分数之间的差。 第二个语音分数的最大值受到关键语音分数的限制。 第二图像分数的最大值受到关键图像分数的限制。
-
公开(公告)号：KR101211918B1
公开(公告)日：2012-12-13
申请号：KR1020100133968
申请日：2010-12-23
Applicant: 전남대학교산학협력단
IPC: G10L17/00
Abstract: 본발명은로그우도스코어기반의신뢰도계산방법에관한것으로, 보다구체적으로는음성및 영상의 신뢰도값을계산하기위한로그우도스코어의두 번째최대값을일정한범위이내로조정하는최적화인자및 상기신뢰도값이 0 내지 1 사이의값으로계산되게하는최적화변수를이용하여, 상기신뢰도값을과대예측또는과소예측되지않고정확한수치로계산되게하는로그우도스코어기반의신뢰도계산방법에관한것이다.
-
公开(公告)号：KR101218666B1
公开(公告)日：2013-01-21
申请号：KR1020100133965
申请日：2010-12-23
Applicant: 전남대학교산학협력단
Abstract: 본 발명은 멀티모달 화자인식 시스템의 인식성능 향상을 위한 입력영상의 신뢰도 계산방법에 관한 것으로, 보다 구체적으로는 영상 및 음성을 이용하여 화자를 인식하는 멀티모달 화자인식 시스템에서 상기 영상과 상기 음성이 각각 입력된 후 통합되기 이전에, 상기 입력되는 영상과, 기저장된 표준영상과의 밝기 차이를 가우시안 퍼지 멤버쉽 함수를 이용하여 수치화한 영상신뢰도를 계산한 후 상기 영상신뢰도를 이용하여 상기 영상에 반영될 영상 가중치를 계산하는 멀티모달 화자인식 시스템의 인식성능 향상을 위한 입력영상의 신뢰도 계산방법에 관한 것이다.
-
公开(公告)号：KR1020120072149A
公开(公告)日：2012-07-03
申请号：KR1020100133965
申请日：2010-12-23
Applicant: 전남대학교산학협력단
CPC classification number: G06K9/6293 , G06K9/6202 , G06T7/44
Abstract: PURPOSE: A method for calculating reliability of an input image to increase recognizing performance of a multimodal speaker recognizing system is provided to calculate an image weight value for determining an image dependant degree using reliability and reflect the image weight value when an inputted image is integrated with a voice. CONSTITUTION: The brightness average value of each standard image is calculated by all brightness average value of the standard images(S1200). The brightness average value of an input image is calculated(S1300). The height of a Gaussian function is set by 1. All brightness values of the standard images become central in the Gaussian function. The brightness values of the standard images are set as a standard deviation in the Gaussian function. A first Gaussian function value about the brightness average value of the input image is calculated by image reliability in the Gaussian function(S1400).
Abstract translation: 目的：提供一种用于计算输入图像的可靠性以增加多模式扬声器识别系统的识别性能的方法，以计算用于使用可靠性确定图像相关度的图像权重值，并且当输入的图像与 一个声音。 构成：通过标准图像的所有亮度平均值计算每个标准图像的亮度平均值（S1200）。 计算输入图像的亮度平均值（S1300）。 高斯函数的高度设置为1.标准图像的所有亮度值在高斯函数中变为中心。 将标准图像的亮度值设置为高斯函数中的标准偏差。 关于输入图像的亮度平均值的第一高斯函数值通过高斯函数中的图像可靠性来计算（S1400）。
-
-
-
Search Results
4
records
(took 0.018 seconds)
