秖秖てvector quantization弧琌ㄓ戈秖程盽砆矗癚阶よ猭ぇぷㄤ琌肚参戈溃罽烩办秖秖て琌ぃ吏
俱秖秖て筁祘い程璶獽琌絪絏codebook砞璸┮孔絪絏琌竤絪絏秖codevector ┪ codeword栋τи程沧ヘ夹獽琌硂ㄇ珼匡ㄓ絪絏秖ㄓ赣丁い场戈秖
羭ㄒㄓ弧瓜┮ボ蝴丁いи盢俱丁だ澄Θ璝单だ–单だ絪絏秖堵翴ㄓㄓ丁い场秖常硂ㄇΤ絪絏秖ㄓ蠢硂獽琌秖秖て弘┮ 瓜5-2-4.a蝴丁秖秖て絛ㄒ
安砞x琌D蝴丁い秖瓜侩醚瑈祘いx碞琘贺疭紉﹃feature stringㄒ紇钩矪瞶い琌DCT玒计粂侩醚い琌LPClinear prediction coefficientи辨硓筁琘贺癸莱よΑQ盢场x癸莱妓琌D蝴丁いΤ絪絏秖yiτyi计ヘ獽嘿赣絪絏
眖芠ㄓ癸莱Qヘ讽礛琌辨秖x癸莱yiア痷distortion dx, yi禫禫τdx, yiΑ妓禯瞒︳代┪琌ア痷︳代よΑㄒキよ畉square error
程膀セ秖秖て簍衡猭莱赣璶衡琌1980Y. Linde, A. Buzo, ㎝R. Gray矗秖秖て簍衡猭 [25]ㄓ碞盢赣簍衡猭嘿LBG簍衡猭LBG簍衡猭セ借蹦ノk-meansだ竤猭 [1]沮箇戳絪絏安砞琌k盢┮Τ癡絤秖training vectorだk竤τ竤竤いみ獽Θ竤絪絏秖
ぇ贺э▆秖秖て簍衡猭ぃ耞砆矗ㄓ [21] [9]膀セ常蹦ノ贺攫だ摸よΑ俱秖秖て筁祘虏てΘ碭˙艼
˙艼いだ澄疭紉玥沮ぃ莱ノのぃ簍衡猭τΤ┮ぃㄒ紇钩矪瞶い獽琌局Τ程跑钵秖DCT玒计 [3] ョ┪琌蹦ノ璶だ秖だ猂猭principal component analysisт耕璶щ紇疭紉 [26]
- 盢场癡絤秖耴Θ竤
- 珼匡琘贺だ澄疭紉盢–攫挡篶程┏糷竤籈だ澄Θㄢ场だ秈︽k-means だ竤猭だ澄よΑパいみ翴秨﹍匡拒猽璶だ秖principal componentよオㄢ翴讽 k-means 币﹍いみ翴
- 安程┏糷竤籈计ヘご礛箇戳絪絏玥˙艼
- 璸衡攫挡篶程┏糷–竤籈竤いみ竤絪絏秖
Data Clustering and Pattern Recognition (戈だ竤籔妓Α侩粄)