First , the theories of the music algorithm and the esprit are presented here . conventional algorithms are limited by the array configuration , and a constructing vectors algorithm , which uses the correlative function of array data , is proposed in this paper . this algorithm is n ' t restricted within the special array configuration , and it is also very steady 在介紹了多重信號分類( music )算法和旋轉(zhuǎn)不變技術(shù)( esprit )的基本原理后,考慮到常規(guī)的算法都受到陣列形式的限制,本文在esprit算法的基礎(chǔ)上,提出了一種利用陣元數(shù)據(jù)的相關(guān)函數(shù)構(gòu)造向量的算法,該算法不要求特定陣列結(jié)構(gòu),且有一定的穩(wěn)健性。
In this paper , we begin with the analysis of wavelet transform . after the analysis of image wavelet coefficients and methods of image compression , a method of vector - constitution among different subbands , making verctor book using pcc + lbg , and fast vq is presented . at the same time a better compression performance is improved by using multistage vector algorithm , the design of this algorithm based on dsps is given at the end of this paper 該算法充分利用了小波分解后各子帶間的相關(guān)性,跨子帶構(gòu)造高維數(shù)矢量,利用改進的漸進構(gòu)造聚類( pcc )結(jié)合lbg的算法生成了具有代表性的最優(yōu)碼書,并提取特征矢量快速實現(xiàn)矢量量化,最后通過二級量化進一步降低矢量量化的復雜度。
Statistical learning theory derives necessary and sufficient conditions for consistency and fast rate of convergence of the empirical risk minimization principle , which is the basis of most traditional learning algorithms . it also theoretically underpins the support vector algorithms . support vector learning algorithm is based on structural risk minimization principle 傳統(tǒng)的學習算法大多是基于經(jīng)驗風險最小化原則的,統(tǒng)計學習理論給出了經(jīng)驗風險最小化原則一致和快速收斂的充分和必要條件,并且為支持向量算法做了理論支持。