运用於影像向量量化的快速预测搜寻演算法精选

《运用於影像向量量化的快速预测搜寻演算法精选》由会员上传分享，免费在线阅读，更多相关内容在应用文档-天天文库。

1、運用於影像向量量化的快速預測搜尋演算法翁旭谷摘　要本文提出一個以小波轉換和加權平均卡曼濾波器為基礎的快速預測搜尋演算法運用於影像的向量量化。使用提出的演算法，最小誤差碼字可以只要搜尋部份的小波轉換碼本中找出。假如最小誤差碼字被發現落於加權平均卡曼濾波演算法所求得的預測搜尋區域中，那麼比由全域搜尋所求得的絕對位址較小的相對位址，將被送至解碼端，如此由模擬的結果發現，本文所提出的方法與全域搜法比較起來品質一樣，但被大大的降低計算量，且降低大約30%的位元率。FastPredictiveSearchAlgorithmforVectorQua

2、ntizationofImagesShiuh-KuWengABSTRACTThepaperpresentsafastpredictivesearchalgorithmforvectorquantization(VQ)basedonwavelettransformandWeightedAverageKalmanFilter(WAKF).Withtheproposedalgorithm,theminimumdistortioncodewordcanbefoundbysearchingonlyaportionofthewavelettran

3、sformedcodebook.IftheminimumdistortioncodewordfoundisfallenintoapredictedsearchareaobtainedbytheWAKFalgorithm,therelativeaddressthatisshorterthantheabsoluteaddressforfullsearchrangeissenttodecoder.Simulationresultsindicatethattheproposedalgorithmachievesasignificantredu

4、ctionincomputationsandabout30%bit-ratereduction,ascomparedtotheconventionalfullsearchVQs.Inaddition,thereconstructedqualityisequivalenttothatofthefullsearchalgorithme.1I、IntroductionVectorquantization(VQ)hasbeenproventobeeffectiveinspeechandimagecompressionduetoitsexcel

5、lentrate-distortionperformance[1]-[3].Theimagetobeencodedisfirstpartitionedintoasetofblocksrepresentedbyvectors.Eachvectoriscomparedwithallthecodewordsinapredesignedcodebook.Thecompressionisachievedbytransmittingtheindexofthebest-matchedcodewordtothereceiverinsteadofthe

6、vectoritself.Thedecoderreconstructstheimageblockaccordingtothereceivedindexbyusingatablelookupprocedure.Thecomparisonofaninputvectorwiththecodewordsinthecodebookiscalledencodingsearch.Theencodingsearchisverycomputationallyintensiveanditscomplexitygrowsexponentiallywitht

7、hebitrateandvectordimension.ThebasicVQyieldsthebitrateintherangeof0.5~0.6bit/pixel(bpp)formonochromeimages[1].TherearetwomainaspectstoimprovethebasicVQ:toraisecompressionefficiencyandtoreduceencodingtimecomplexity.Intheliterature,thetwoaspectsareoftenconsideredindepende

8、ntly.Inordertoimprovecompressionefficiency,variousmemoryVQtechniquesthatexploitthecorrelationofneighborblocksh