基于Web圖像的Kinship關(guān)系驗(yàn)證研究
發(fā)布時(shí)間：2018-08-18 08:07
【摘要】：計(jì)算機(jī)視覺(jué)系統(tǒng)的最終目標(biāo)是要獲得自適應(yīng)能力、自學(xué)習(xí)能力、在各種解決方案中權(quán)衡的能力、對(duì)新的上下文情景和應(yīng)用場(chǎng)合進(jìn)行泛化的能力,以及和其它系統(tǒng)(包括人)進(jìn)行溝通的能力。人臉作為計(jì)算機(jī)視覺(jué)領(lǐng)域中的一種重要研究對(duì)象,因其在圖像獲取過(guò)程中的便利性和廉價(jià)性,受到了模式識(shí)別和機(jī)器學(xué)習(xí)等相關(guān)領(lǐng)域科研人員的廣泛關(guān)注,經(jīng)過(guò)近三十年的發(fā)展,人臉識(shí)別系統(tǒng)已經(jīng)開(kāi)始從實(shí)驗(yàn)室進(jìn)入商業(yè)領(lǐng)域。然而,在這一從實(shí)驗(yàn)室到具體應(yīng)用場(chǎng)景的轉(zhuǎn)移過(guò)程中,存在多種不同性質(zhì)的人臉識(shí)別問(wèn)題,其中一些還非常困難,例如對(duì)從網(wǎng)頁(yè)中采集的人臉圖像進(jìn)行親屬關(guān)系驗(yàn)證的問(wèn)題。基于人臉圖像進(jìn)行親屬關(guān)系驗(yàn)證面臨新的問(wèn)題和挑戰(zhàn),這些問(wèn)題主要來(lái)自于組圖像的表示和驗(yàn)證器的設(shè)計(jì)兩個(gè)方面。其中組圖像表示方面的問(wèn)題包括由成像環(huán)境、表情、遮擋、姿態(tài)和遺傳特性等造成的人臉外觀上的豐富變化。而驗(yàn)證器的設(shè)計(jì)則面臨組圖像刻畫(huà)困難、目標(biāo)類(lèi)信息缺失和遺傳差異大等因素。正是由于這些挑戰(zhàn)存在,使得之前的人臉驗(yàn)證算法難以直接被用于處理親屬關(guān)系驗(yàn)證,急需研究新的解決方案來(lái)應(yīng)對(duì)這些問(wèn)題。本文重點(diǎn)研究了基于Web圖像的魯棒的親屬關(guān)系驗(yàn)證問(wèn)題。本文重點(diǎn)討論親屬關(guān)系驗(yàn)證中涉及的三個(gè)核心問(wèn)題,即親屬關(guān)系主體對(duì)象的表示學(xué)習(xí),親屬關(guān)系驗(yàn)證器的設(shè)計(jì)和在實(shí)際應(yīng)用場(chǎng)合中的推廣。針對(duì)第一個(gè)核心問(wèn)題,提出了一種基于軟投票的親屬關(guān)系人臉特征塊的選擇算法;針對(duì)第二個(gè)問(wèn)題,探討了嵌入一定先驗(yàn)信息的組親屬關(guān)系驗(yàn)證模型;針對(duì)在實(shí)際應(yīng)用場(chǎng)合中的推廣,提出了混合親屬關(guān)系驗(yàn)證問(wèn)題及其模型設(shè)計(jì)方法。具體地,本文的主要貢獻(xiàn)和創(chuàng)新點(diǎn)可以總結(jié)為如下幾點(diǎn):(1)提出一種考慮組關(guān)系的親屬關(guān)系驗(yàn)證問(wèn)題并發(fā)布一個(gè)包含超過(guò)1000組家庭的親屬關(guān)系人臉數(shù)據(jù)集。親屬關(guān)系驗(yàn)證學(xué)習(xí)可以被看作是向刻畫(huà)多個(gè)視覺(jué)對(duì)象之間互信息的邁進(jìn),然而已有的親屬關(guān)系驗(yàn)證研究大多考慮的是對(duì)關(guān)系,即父—子,父—女,母—子和母—女關(guān)系,但在實(shí)際應(yīng)用領(lǐng)域,親屬關(guān)系包括更加復(fù)雜的主體關(guān)系,而在所有人類(lèi)社會(huì)關(guān)系中的核心基礎(chǔ)單元是父母—兒子和父母—女兒家庭關(guān)系,理解該親屬關(guān)系將促進(jìn)人工智能對(duì)人類(lèi)社會(huì)行為的理解,也是實(shí)現(xiàn)計(jì)算機(jī)視覺(jué)系統(tǒng)從對(duì)單一對(duì)象的刻畫(huà)到多個(gè)主體對(duì)象描述的飛躍,另外,相較于更復(fù)雜的親屬關(guān)系驗(yàn)證,組親屬關(guān)系驗(yàn)證更容易實(shí)現(xiàn),因?yàn)槠渖婕暗姆懂犑强煽氐?且問(wèn)題本身也更容易定義。(2)提出了一種基于軟投票的親屬關(guān)系人臉特征塊選擇方法。探討了基于有監(jiān)督方式的親屬關(guān)系表示學(xué)習(xí),實(shí)現(xiàn)親屬關(guān)系特征提取的判別性和魯棒性。主要針對(duì)現(xiàn)有親屬關(guān)系表示學(xué)習(xí)僅使用家庭主體中的某個(gè)單一對(duì)象,而親屬關(guān)系主體之間又具有一定空間結(jié)構(gòu)關(guān)系問(wèn)題,考慮挖掘主體對(duì)象之間的相關(guān)性并利用這些相關(guān)性探尋親屬主體間的判別信息。具體地,在給定圖像中每個(gè)位置上的所有單個(gè)特征完成相互之間的競(jìng)爭(zhēng)后,再選擇一些圖像組,而這些組所包含的獲勝單特征的比例更高。該方法的主要優(yōu)點(diǎn)是相較于主流的人臉局部特征選擇算法更加靈活,因?yàn)槠涫窃谝环N更加精細(xì)的級(jí)別進(jìn)行特征選擇,因此可以獲得更高的性能。(3)提出了一種嵌入人類(lèi)社會(huì)學(xué)知識(shí)的相對(duì)對(duì)稱(chēng)的組親屬關(guān)系驗(yàn)證模型。考慮到現(xiàn)有親屬關(guān)系驗(yàn)證必須要面對(duì)的問(wèn)題,即小樣本問(wèn)題,而借助額外判別信息又是解決小樣本問(wèn)題的一個(gè)有力手段,受人類(lèi)社會(huì)學(xué)研究成果的啟發(fā),將孩子和父母中某一方較為相似的先驗(yàn)信息嵌入模型,提出一種相對(duì)對(duì)稱(chēng)雙線性模型,在TSKinFace和KinFaceW親屬關(guān)系人臉數(shù)據(jù)集上驗(yàn)證了算法的有效性。另外,當(dāng)父母雙方信息均已知時(shí),該方法還可用于解決對(duì)親屬關(guān)系驗(yàn)證問(wèn)題,相對(duì)于基于父母中一方進(jìn)行判定的方法具有較好的推廣性,一定程度上彌補(bǔ)了待驗(yàn)證人臉的身份信息,在TSKinFace數(shù)據(jù)集上驗(yàn)證了算法的有效性。最后,所提方法可以被看作為一個(gè)框架,在該框架中可以通過(guò)有效地嵌入先驗(yàn)信息的方式整合任何一種用于處理對(duì)親屬關(guān)系驗(yàn)證的方法來(lái)應(yīng)對(duì)組親屬關(guān)系驗(yàn)證問(wèn)題。(4)提出了一種混合親屬關(guān)系驗(yàn)證問(wèn)題及其模型設(shè)計(jì)方法。主要針對(duì)現(xiàn)有親屬關(guān)系驗(yàn)證都是基于給定主體的性別種類(lèi)分別進(jìn)行研究而為實(shí)際應(yīng)用帶來(lái)額外的性別標(biāo)注工作量的問(wèn)題,探討了親屬關(guān)系驗(yàn)證模型在實(shí)際應(yīng)用場(chǎng)景中的推廣,提出了混合對(duì)親屬關(guān)系驗(yàn)證。具體的,受人類(lèi)社會(huì)學(xué)研究成果的啟發(fā),即一些人臉外觀,如眼睛、頭發(fā)顏色、酒窩、皮膚等表現(xiàn)出極強(qiáng)的遺傳性,將不同親屬關(guān)系看作為不同但相互之間有相關(guān)性的任務(wù),并使用多任務(wù)學(xué)習(xí)框架將每個(gè)任務(wù)模型分解為兩個(gè)部分,即一部分在所有任務(wù)間共享,另一部分則被每種任務(wù)獨(dú)享。這兩部分在一個(gè)聯(lián)合框架下同時(shí)學(xué)習(xí),使得所提算法能利用到多個(gè)任務(wù)之間的共有信息。另外,該方法的優(yōu)點(diǎn)是,當(dāng)每個(gè)任務(wù)僅有很少訓(xùn)練樣本時(shí),能通過(guò)在任務(wù)間遷移信息的手段互補(bǔ)判別性信息以達(dá)到提高算法泛化性的目的。進(jìn)一步,為了使算法更加魯棒,提出了一個(gè)多視圖多任務(wù)的混合對(duì)親屬關(guān)系驗(yàn)證模型,其中通過(guò)為不同的特征學(xué)習(xí)各自不同的權(quán)重融合多種特征以提高混合對(duì)親屬關(guān)系驗(yàn)證的性能。
[Abstract]:The ultimate goal of computer vision systems is to acquire the ability of self-adaptation, self-learning, the ability to weigh among solutions, the ability to generalize new contexts and applications, and the ability to communicate with other systems (including people). Because of its convenience and low cost in the process of image acquisition, it has attracted extensive attention of researchers in the fields of pattern recognition and machine learning. After nearly 30 years of development, face recognition system has begun to enter the commercial field from the laboratory. However, in the process of the transition from the laboratory to the specific application scenario, there exists a lot of problems. There are many different kinds of face recognition problems, some of which are still very difficult, such as the problem of kinship verification of face images collected from web pages. The problems of group image representation include the rich changes of facial appearance caused by imaging environment, expression, occlusion, posture and genetic characteristics. The design of the validator is faced with the difficulties of group image description, target class information missing and genetic differences. This paper focuses on the robust relational validation problem based on Web images. This paper focuses on three core issues involved in relational validation, namely, representation learning of relational subject objects and relational validator. Aiming at the first core problem, this paper proposes an algorithm for selecting the feature blocks of relatives based on soft voting; for the second problem, a group relatives validation model embedding certain prior information is discussed; for the promotion in practical application, a hybrid relatives algorithm is proposed. Specifically, the main contributions and innovations of this paper can be summarized as follows: (1) A relational validation problem considering group relationships is proposed and a relational face dataset containing more than 1000 families is published. Mutual information advances between visual objects, however, most of the existing kinship validation studies have considered pairwise relationships, i.e. father-son, father-daughter, mother-son and mother-daughter relationships. In practical applications, kinship includes more complex subject relationships, and the core unit of all human social relationships is parent-son. Understanding a parent-daughter family relationship will facilitate AI's understanding of human social behavior, as well as a leap in computer vision systems from depicting a single object to describing multiple subject objects. In addition, group kinship validation is easier to implement than more complex kinship validation because of its involvement. (2) A method of feature block selection based on soft voting is proposed for relational facial feature extraction. The method is based on supervised relational representation learning to realize the discriminability and robustness of relational feature extraction. A single object in a family subject, and the relatives have a certain spatial structure relationship between them. Considering mining the relativity between the subject objects and exploring the discriminant information between the relatives, all the individual features in each position in a given image compete with each other. The main advantage of this method is that it is more flexible than the mainstream face feature selection algorithm, because it is a more fine level of feature selection, so it can obtain higher performance. (3) A new embedded human face feature selection algorithm is proposed. Relatively symmetrical group relational validation model of sociological knowledge. Considering the existing problems in relational validation, that is, the small sample problem, and the use of additional discriminant information is a powerful means to solve the small sample problem, inspired by the results of anthropological sociology, children and one of the parents are more similar. A priori information embedding model is proposed, and a relative symmetric bilinear model is proposed to verify the validity of the proposed algorithm on TSKinFace and KinFaceW relational face datasets. In addition, when both parents'information is known, this method can also be used to solve the problem of relational validation. Finally, the proposed method can be regarded as a framework in which any method used to process relational validation can be integrated by effectively embedding prior information. (4) Propose a hybrid kinship validation problem and its model design method. Mainly aim at the problem that the existing kinship validation is based on the gender type of given subject and brings extra gender labeling workload for practical application, and discuss the kinship validation model. A hybrid approach is proposed to validate kinship in practical scenarios. Specifically, inspired by anthropological research, some facial features, such as eyes, hair color, dimples, and skin, exhibit strong inheritance. Different kinship relationships are viewed as different but related tasks and are used widely. Task learning framework decomposes each task model into two parts, one shared by all tasks and the other shared by each task. The two parts learn simultaneously in a joint framework, enabling the proposed algorithm to take advantage of the common information between multiple tasks. Furthermore, in order to make the algorithm more robust, a multi-view and multi-task hybrid pairwise kinship verification model is proposed, in which different weights are fused by learning for different features. Features to enhance the performance of hybrid validation for kinship.
【學(xué)位授予單位】：南京航空航天大學(xué)
【學(xué)位級(jí)別】：博士
【學(xué)位授予年份】：2016
【分類(lèi)號(hào)】：TP391.41
本文編號(hào)：2188873
[Abstract]:The ultimate goal of computer vision systems is to acquire the ability of self-adaptation, self-learning, the ability to weigh among solutions, the ability to generalize new contexts and applications, and the ability to communicate with other systems (including people). Because of its convenience and low cost in the process of image acquisition, it has attracted extensive attention of researchers in the fields of pattern recognition and machine learning. After nearly 30 years of development, face recognition system has begun to enter the commercial field from the laboratory. However, in the process of the transition from the laboratory to the specific application scenario, there exists a lot of problems. There are many different kinds of face recognition problems, some of which are still very difficult, such as the problem of kinship verification of face images collected from web pages. The problems of group image representation include the rich changes of facial appearance caused by imaging environment, expression, occlusion, posture and genetic characteristics. The design of the validator is faced with the difficulties of group image description, target class information missing and genetic differences. This paper focuses on the robust relational validation problem based on Web images. This paper focuses on three core issues involved in relational validation, namely, representation learning of relational subject objects and relational validator. Aiming at the first core problem, this paper proposes an algorithm for selecting the feature blocks of relatives based on soft voting; for the second problem, a group relatives validation model embedding certain prior information is discussed; for the promotion in practical application, a hybrid relatives algorithm is proposed. Specifically, the main contributions and innovations of this paper can be summarized as follows: (1) A relational validation problem considering group relationships is proposed and a relational face dataset containing more than 1000 families is published. Mutual information advances between visual objects, however, most of the existing kinship validation studies have considered pairwise relationships, i.e. father-son, father-daughter, mother-son and mother-daughter relationships. In practical applications, kinship includes more complex subject relationships, and the core unit of all human social relationships is parent-son. Understanding a parent-daughter family relationship will facilitate AI's understanding of human social behavior, as well as a leap in computer vision systems from depicting a single object to describing multiple subject objects. In addition, group kinship validation is easier to implement than more complex kinship validation because of its involvement. (2) A method of feature block selection based on soft voting is proposed for relational facial feature extraction. The method is based on supervised relational representation learning to realize the discriminability and robustness of relational feature extraction. A single object in a family subject, and the relatives have a certain spatial structure relationship between them. Considering mining the relativity between the subject objects and exploring the discriminant information between the relatives, all the individual features in each position in a given image compete with each other. The main advantage of this method is that it is more flexible than the mainstream face feature selection algorithm, because it is a more fine level of feature selection, so it can obtain higher performance. (3) A new embedded human face feature selection algorithm is proposed. Relatively symmetrical group relational validation model of sociological knowledge. Considering the existing problems in relational validation, that is, the small sample problem, and the use of additional discriminant information is a powerful means to solve the small sample problem, inspired by the results of anthropological sociology, children and one of the parents are more similar. A priori information embedding model is proposed, and a relative symmetric bilinear model is proposed to verify the validity of the proposed algorithm on TSKinFace and KinFaceW relational face datasets. In addition, when both parents'information is known, this method can also be used to solve the problem of relational validation. Finally, the proposed method can be regarded as a framework in which any method used to process relational validation can be integrated by effectively embedding prior information. (4) Propose a hybrid kinship validation problem and its model design method. Mainly aim at the problem that the existing kinship validation is based on the gender type of given subject and brings extra gender labeling workload for practical application, and discuss the kinship validation model. A hybrid approach is proposed to validate kinship in practical scenarios. Specifically, inspired by anthropological research, some facial features, such as eyes, hair color, dimples, and skin, exhibit strong inheritance. Different kinship relationships are viewed as different but related tasks and are used widely. Task learning framework decomposes each task model into two parts, one shared by all tasks and the other shared by each task. The two parts learn simultaneously in a joint framework, enabling the proposed algorithm to take advantage of the common information between multiple tasks. Furthermore, in order to make the algorithm more robust, a multi-view and multi-task hybrid pairwise kinship verification model is proposed, in which different weights are fused by learning for different features. Features to enhance the performance of hybrid validation for kinship.
【學(xué)位授予單位】：南京航空航天大學(xué)
【學(xué)位級(jí)別】：博士
【學(xué)位授予年份】：2016
【分類(lèi)號(hào)】：TP391.41
【相似文獻(xiàn)】
相關(guān)期刊論文 前2條
1 郭瑞強(qiáng);閆紹惠;趙書(shū)良;申玉鳳;;親屬關(guān)系網(wǎng)絡(luò)的關(guān)系追溯算法[J];計(jì)算機(jī)應(yīng)用;2014年07期
2 程美寶;;網(wǎng)上織網(wǎng):當(dāng)代親屬關(guān)系的建構(gòu)[J];學(xué)術(shù)研究;2008年09期
相關(guān)重要報(bào)紙文章 前7條
1 華中師范大學(xué)政治學(xué)研究院 吳理財(cái);鄉(xiāng)村社會(huì)個(gè)體化的一種反應(yīng)[N];社會(huì)科學(xué)報(bào);2014年
2 早報(bào)見(jiàn)習(xí)記者 陳竹沁;“王曉玲與兩股東非親屬關(guān)系”[N];東方早報(bào);2014年
3 特約記者 云也退;人的身份應(yīng)該是復(fù)雜多變的[N];第一財(cái)經(jīng)日?qǐng)?bào);2014年
4 本報(bào)記者 杜昌營(yíng);親屬關(guān)系證明等事項(xiàng)可網(wǎng)上申辦[N];人民公安報(bào);2014年
5 東方;香港接納內(nèi)地移民擬實(shí)行新計(jì)分制度[N];法制日?qǐng)?bào);2002年
6 ;未成年人監(jiān)護(hù)應(yīng)有說(shuō)法[N];中國(guó)婦女報(bào);2000年
7 李 華 馮劍曉;“醫(yī)學(xué)上認(rèn)為不能結(jié)婚的疾病”應(yīng)該明確[N];人民法院報(bào);2004年
相關(guān)博士學(xué)位論文 前2條
1 秦曉倩;基于Web圖像的Kinship關(guān)系驗(yàn)證研究[D];南京航空航天大學(xué);2016年
2 王利芬;延續(xù)、疏遠(yuǎn)和理性化[D];華中科技大學(xué);2015年
相關(guān)碩士學(xué)位論文 前10條
1 李應(yīng)芝;彝族擬親屬關(guān)系研究[D];西南大學(xué);2012年
2 黃巖;農(nóng)村擬親屬關(guān)系研究[D];濟(jì)南大學(xué);2012年
3 譚建榮;親屬關(guān)系影響定罪量刑研究[D];西南政法大學(xué);2013年
4 滿昌朋;個(gè)體化、親屬關(guān)系與老年人精神贍養(yǎng)研究[D];西南交通大學(xué);2015年
5 童欣弘;已婚子女與父母的互動(dòng)傾向研究[D];杭州師范大學(xué);2016年
6 張浩;基于親屬關(guān)系網(wǎng)絡(luò)的圖模式匹配方法研究[D];河北師范大學(xué);2016年
7 張霞;基于親屬關(guān)系網(wǎng)絡(luò)的子圖查詢方法研究[D];河北師范大學(xué);2017年
8 李虎;壯族擬親屬關(guān)系的研究[D];廈門(mén)大學(xué);2008年
9 嚴(yán)子文;親屬雇傭中的糾紛與沖[D];吉林大學(xué);2013年
10 閆紹惠;親屬關(guān)系網(wǎng)絡(luò)的結(jié)構(gòu)分析及關(guān)系追溯算法研究[D];河北師范大學(xué);2014年
，本文編號(hào)：2188873
本文鏈接：http://sikaile.net/shoufeilunwen/xxkjbs/2188873.html
最近更新
教材專(zhuān)著
