201137651 六、發明說明： 【發明所屬之技術領域】 本發明，係有關於具備資訊過濾檢測功能的記憶體、 其使用方法、包含該記憶體的裝置。 【先前技術】 在資訊被作資料化並能夠方便地作利用之時代中，爲 了能夠從此龐大之資訊資料中而檢測出適當的資訊並作利 用，係仍殘留有各種的課題。 特別是，關於在以畫像辨識、聲音辨識、OCR文字辨 識、全文檢索、指紋等之生物體認證等作爲代表之資訊檢 測中而成爲共通之基礎的技術中，係存在有從資訊中而將 一致或者是類似的資訊（圖案）檢測出來或者是解析出來 之圖案辨識技術’此種技術，在社會基礎公共設備、產業 用設備、工廠設備或者是數位相機或家電產品、乃至於最 新的機器人或者是人工智慧等之各種技術領域中，均有被 作利用’並成爲在高度之資訊處理中所不可或缺者。 然而’在以圖案辨識作爲其中一例之資訊檢測之技術 上最大之課題’係在於資訊比較時之比較組合次數（檢索 次數），通常’係尋找出對象資訊中的最適演算法來將組 合比較次數（檢索次數）作削減、或者是依存於作爲對象 之資訊的內容而使用超級電腦等之高速的演算處理機器來 尋找出答案，又，視情況，亦會有需要對於檢測之精確度 作犧牲的情形。 -5- 201137651 本發明，係爲實現一種能夠對於上述一般之身爲圖案 辨識或者是資訊檢測技術中的長年課題之資訊檢測精確度 作保證並且能夠將比較組合次數（檢索次數、檢索時間） 作最大限度之降低的記憶體、以及該記憶體之利用方法， 並爲對於由相同申請人、相同發明人在20 10年2月18曰所 提出的日本專利申請案特願201 0-33376之「資訊處理裝置 之資訊的共通管理方法、資訊之檢測方法、資料以及位址 之相對關係總括並列比較聯想記憶體、具備有將資訊作共 通管理之功能的資訊處理裝置、其之軟體程式」中的與「 資料以及位址之相對關係總括並列聯想記憶體」有所關連 之全部項目，而對本申請案主張優先權者。 同樣地，2010年3月4日的日本專利申請案特願2010-472 1 5「具備有資訊過濾檢測功能之半導體積體電路、其 之使用方法、使用此半導體積體電路之裝置」，係爲從上 述日本特願20 1 0-3 3 3 76作了獨立的發明，並將發明名稱之 表現變更爲「具備有資訊過濾檢測功能之半導體積體電路 j ，而對於本發明之最大特徵的「二重並列之合格與否判 定結果之邏輯演算的想法」更加明確化者，又，係將作檢 測之資訊的範圍，並不僅侷限於二維之畫像，而亦擴大成 從一維至多維之資訊者，並將此一申請案全體，主張爲對 於本案之優先權。 本申請案，係將發明名稱明確地表現爲「具備資訊過 濾檢測功能的記憶體」，並將上述2件前申請案作了統合 ，而以上述日本特願20 1 0-472 1 5爲主，再追加用以削減資 201137651 訊過濾電路數之手段、二重並列邏輯演算之多重化手段、 乃至於以對於人工智慧之應用等作爲其中一例之此記憶體 的使用方法’而補足說明之不足處，並對於一部份之表現 方法作了變更。 如同前述所說明一般，以圖案辨識或者是圖案匹配作 爲其中一例之資訊檢測的技術，其範圍係極爲廣泛，在與 該檢索時間之縮短有所關連的發明中，係存在有龐大的數 量’但是，像是本發明一般之爲了縮短檢測時間而將身爲 馮諾依曼（Von Neumann )型電腦之宿命的「個別記憶體 之逐次處理」一事本質性地作了避免的手法或者是此種記 憶體之例，係並不存在。 作爲參考，在日本特開平7-1 145：77之「資料檢索裝置 、資料壓縮裝置以及方法」中，係揭示有一種用以藉由相 鄰接之資訊彼此間的比較而對於資訊作反覆檢索之手法， 但是，本申請案之發明，係並不僅是對於相鄰接之資訊彼 此作比較，而是亦以全記憶體之資訊作爲對象而將資料之 內容與其之位址的位置關係作了二重並列比較者。 [先前技術文獻] [專利文獻] [專利文獻1]曰本特開平7-114577號公報 【發明內容】 [發明所欲解決之課題] 201137651 本發明所欲解決之課題，係在於：在將以一維乃至多 維地而作了位址配列之資訊或者是能夠以一維乃至多維地 而作位址配列之資訊作爲對象，並對於作檢測之資訊（未 知之資訊）和成爲檢測之基準的資訊（既知之資訊）的相 互之資訊的複數之位址，而藉由該資料和其之位址的雙方 之關係均相對於條件而成爲合格者，來從未知之資訊中而 判定出與既知之資訊爲相同資訊或者是類似資訊者一般之 圖案辨識或者是知識處理等的資訊檢測中，而實現一種能 夠對於檢測之精確度作保證，並將資料之比較次數作最大 限度之削減，並且使記憶體本身能夠進行上述資訊之檢測 的非馮諾依曼型之資訊檢測記憶體、並且確立該記憶體之 使用方法。 [用以解決課題之手段] 爲了解決上述課題，在申請項1中，係記載有一種將 資訊記憶在每一記憶體位址中並且能夠將該資訊作讀出之 記憶體，其特徵爲，該記憶體係具備有以下（1 )至（4 ) :(1 )用以將從外部所賦予並用以和被記憶在此記憶體 中之資料並列地作比較之第1比較資料、和用以與此記憶 體之位址的位址彼此間作並列比較之第2比較資料，此些 之各比較資料作輸入之輸入手段；和（2)藉由第1比較資 料而和被記憶在此記憶體中之資料作並列比較並對於合致 與否作判定之手段；和（3 )藉由第2比較資料而和此記億 體之位址彼此作並列比較並對於合致與否作判定之手段； -8- 201137651 和（4 )將以上（2 ) 、 （ 3 )雙方之合致與否的判定結果 在每一位址處而並列地作邏輯演算之資料與位址的各合致 與否結果之邏輯演算手段。 在申請項2中’係具備有下述特徵：亦即是，前述具 備資訊過濾檢測功能的記憶體之前述邏輯演算，係爲邏輯 積（AND)演算，並且，係具備有以下（1)至（3)：( 1 )在初次之資訊檢測時，藉由前述第〗比較資料來和記憶 體之資料並列地進行合格與否之判定，並將合格了的至少 1個以上之記憶體位址作爲1次突破位址而作記憶之手段； 和（2 )在下一次以後之資訊檢測時，藉由新的前述第]比 較資料來和記憶體之資料並列地進行合格與否之判定，並 將合格了的記憶體位址，藉由由前述第2比較資料所進行 之位址置換手段來作了置換之位址；和（3)將突破了上 述（1) 、（2)之位址的邏輯積（AND)演算之位址作輸 出之手段。 在申請項3中，係具備有下述特徵：亦即是，前述具 備資訊過濾檢測功能的記憶體，係具備有以下（1 )至（4 ):(1 )對於各記憶體位址而個別地對於由反覆作了特 定次數之賦予的前述各比較資料所進行之比較合格與否判 定結果作計測之計數器手段；和（2 )在前述初次之資訊 檢測時，將合格了的記憶體位址之上述計數器作1的正數 (count up)，並將此作爲目U述1次突破位址之手段；和（ 3 )在上述被反覆作了賦予之前述下一次以後的資訊檢測 時，藉由前述位址之置換手段來將前述邏輯積（AND)演 201137651 算結果在上述1次突破位址之計數器處進行累積正數並作 爲N(2以上之比較次數）次突破位址之手段；和（4)將 上述（3 )之N次突破位址的位址作輸出之手段。 在申請項4中，係具備有下述特徵：亦即是，前述位 址之置換手段，係藉由前述第2比較資料而以位址全範圍 來並列地進行位址之置換。 在申請項5中，係具備有下述特徵：亦即是，身爲前 述第2比較資料之用以與記億體之位址的位址彼此並列地 作比較之資料，係爲下述（1) 、 （2) 2者中之其中一者 的資訊檢測用比較資料：將前述1次突破位址作爲基準位 址’並在前述下一次以後的資訊檢測時，對於藉由前述第 1比較資料而和記億體之資料並列地進行合格與否判定並 合格了的記憶體位址、和上述基準位址，此雙方之位址彼 此的相對位置’而用以判定（1)是否一致的比較資料； (2)是否存在於範圍內之比較資料。 在申請項6中，係具備有下述特徵：亦即是，前述具 備資訊過濾檢測功能的記憶體，係爲將下述（1 )〜（5 ) 中之至少一者的資訊之檢測作爲對象的記憶體構成：（1 )以聲音資訊作爲其中一例之作爲一維資訊而被作記憶或 者是可作記憶之資訊；（2 )以畫像資訊作爲其中一例之 作爲二維資訊而被作記憶或者是可作記憶之資訊；（3 ) 以立體資訊作爲其中一例之作爲三維資訊而被作記憶或者 是可作記憶之資訊；（4 )以時空間資訊作爲其中一例之 作爲多維資訊而被作記憶或者是可作記憶之資訊；（5 ) -10- 201137651 以叢集資訊作爲其中一例之使資訊以位址之群組別而被作 記憶或者是可作記憶之資訊。 在申請項7中，係具備有下述特徵：亦即是，身爲前 述第1比較資料之用以與被記憶在記憶體中之資料作並列 比較的資料，係爲下述（1)〜（5)中之至少1個的資訊 檢測用比較資料：（1 )記憶體資料之一致檢測；（2 )記 憶體資料之大小檢測；（3 )記億體資料之範圍檢測；（4 )記憶體bit個別之比較檢測；（5 ) 3値記憶體資料之比較 檢測。 在申請項8中，係具備有下述特徵：亦即是，前述第1 比較資料、第2比較資料，係藉由下述（1 ) 、 （ 2 )之其 中一者或者是雙方之輸入手段而被作輸入：（1)資料匯 流排；（2 )專用輸入。 在申請項9中，係具備有下述特徵：亦即是，將申請 專利範圍第2項所記載之邏輯積（AND )演算突破或者是 申請專利範圍第3項所記載之前述N次突破位址的位址作輸 出之手段，係藉由下述（1) 、（2)之其中一者或者是雙 方之輸出手段而被作輸出：（1 )資料匯流排；（3 )專用 輸出。 在申請項1 〇中，係具備有下述特徵：亦即是，係在前 述計數器手段中，附加將前述初次資訊檢測時之前述1次 突破位址的位址作記憶之手段，並將對於記憶體位址而個 別進行計測之計數器手段的數量作了削減（計數器手段= 位址數/ η，η係爲自然數）。 -11 - 201137651 在申請項11中，係具備有下述特徵：亦即是，係在前 述具備資訊過濾檢測功能的記憶體中搭載處理器，並n由 上述處理器來實現上述位址之置換手段。 在申請項12中，係具備有下述特徵：亦即是，係具備 有將記憶體之範圍作分割地來實施（2)〜（4)之各個手 段的記憶庫（memory bank)分割比較手段。 在申請項13中，係具備有下述特徵：亦即是，係將甲 請項1所記載之（1 )〜（4 )之各個的手段並列地而具備 有複數。 在申請項Μ中，係具備有下述特徵：亦即是，前述具 備資訊過濾檢測功能的記憶體，係被組入至以CPU作爲其 中一例之其他目的的半導體中並被作使用。 在申請項15中，係具備有下述特徵：亦即是，係在如 申請專利範圍第2項所記載之具備資訊過濾檢測功能的記 憶體中，在前述初次之資訊檢測以及前述下一次以後之資 訊檢測時，根據既知之資訊而將前述各比較資料賦予至此 記憶體中，並藉由讀出前述邏輯積（AND )演算突破位址 ，而從被記憶在此記憶體中之資訊來作資訊檢測出與上述 既知之資訊相同或者是類似的資訊。 在申請項1 6中，係具備有下述特徵··亦即是，係在如 申請專利範圍第3項所記載之具備資訊過濾檢測功能的記 憶體中’將爲了檢測出前述相同資訊或者是前述類似資訊 所需要的充分之複數個數的成爲比較樣本之前述各比較資 料，反覆賦予至此記憶體處，並藉由讀出前述N次突破計 -12- 201137651 數器之計數値，而從被記憶在此記憶體之資訊之中來作資 訊檢測出與上述既知之資訊相同或者是類似的資訊。 在申請項1 7中，係具備有下述特徵：亦即是，在將前 述比較資料抽出時，係求取出相鄰接之樣本間的資料之相 互間的資料差之絕對値，並將藉由對此作集計所得到的樣 本特徵量作爲特定値以上而進行資訊檢測。 在申請項1 8中，係具備有下述特徵：亦即是，與前述 既知之資訊相同或者是類似的資訊，係爲圖案（pattren ) 資訊。 在申請項1 9中’係具備有下述特徵：亦即是，在決定 前述1次突破位置之前述初次資訊檢測時的賦予前述第丨比 較資料之步驟中，係進行下述之其中一者或者是雙方：從 複數種類之樣本來選擇出最適當者而作爲最初之前述比較 樣本’並進行資訊之檢測，或者是使第1比較資料具備有 一定之資料的範圍，來進行資訊之檢測。 在申請項20中’係具備有下述特徵：亦即是，係—倂 使用有能夠對於前述具備資訊過濾檢測功能的記憶體作存 取並進行資料之讀出與寫入的C P U，來進行資訊檢測。 在申請項2 1中’係具備有下述特徵：亦即是，係在前 述具備資訊過濾檢測功能的記憶體中，將前述圖案資訊作 爲知識資訊而作記憶’並進行由檢測出圖案資訊所進行之 知識處理。 在申請項2 2中’係記載有一種裝置，其特徵爲：係包 含有如申請專利範圍第1項所記載之具備資訊過濾檢測功 -13- 201137651 能的記憶體。 [發明之效果] 若依據本發明’則能夠作爲智慧性之具備有知識的記 憶體來作利用，並且’不僅是能夠將所有之資訊的相同性 以及類似性確實且高度地檢測或者是解析出來，亦能夠在 資訊預測之技術領域或者是高度性之知識處理中而作廣泛 之利用，而能夠期待一種由真正的非馮諾依曼型資訊處理 所致的嶄新之資訊處理的流程。 【實施方式】 首先，以身爲二維資訊之畫像爲例，來對於本發明之 資訊檢測的槪念作說明。 通常當藉由檢測對象之畫像（未知之資訊）與檢測基 準之畫像（既知之資訊）來檢測畫像之同一性的情況時， 基本上，係以從成爲檢測基準的畫像（既知之資訊）所採 取出的某些畫像資訊爲基礎，來對於未知之檢測對象的畫 像以全符合方式來作檢索，而在對此要求精確度的情況時 ，係需要該畫像之每一座標。 作爲在檢索中所耗費的時間之其中一例’當將個人電 腦或者是由數位電視訊號所致之電視畫面的特定像素作爲 對象，並從被顯示之畫像上尋找出特定資料的情況時’其 對象係爲200萬像素左右》 當先暫時將此一全畫面之位元映像資料從圖形記億體 -14- 201137651 作資料展開至檢索用記憶體中，並將暫時已展開的資料之 全範圍’使c P U以例如毎像素平均5 0 η秒的速度來進行尋 找出特定資料等單純性檢索的情況時，在初次之全圖形範 圍（全畫面範圍）的檢索係會耗費200萬χ50η秒=l〇〇m秒 ’雖然’通常在第2次以後，由於檢索對象係被作過濾， 因此檢索時間會變短，但是，爲了特定出目的畫像，係會 需要數百m秒左右的時間，故而，當有必要在1個畫面上對 於大量之畫像作檢索的情況時，不論如何地進行高速處理 ，均變得無法忽視檢索時間。 此外’以上之說明，只是完全爲同一畫像的情況，當 假設存在有畫像之尺寸變更或者是旋轉的情況時，由於係 需要反覆實施座標變換之演算，因此亦會有處理時間成爲 上述情況之數百倍乃至數千倍甚至是更長的情況，此種檢 索，係難以實現。 上述情況，可以說是由需要進行記憶體之每一位址的 CPU逐次檢索之馮諾依曼型電腦所進行的資訊檢測之不可 避免的宿命。 由於上述一般之檢索時間的技術性背景，現在之檢索 的主流技術，係成爲由將畫像之特徵作了抽出的特徵資料 之叢集化所致的以畫像彼此間之類似性作爲對象的檢索， 並被利用在最近之以數位相機的臉孔辨識或者是笑臉辨識 等爲首乃至於聲音辨識等之廣範圍的領域中。 然而，像是檢測之精確度或是檢索之時間、所能夠檢 測出之資訊等的檢索能力，係會依存於此些特徵抽出之手 -15- 201137651 法或者是叢集化之手法而大幅度的改變。 又，在畫像之檢索的利用領域中，亦多會有誤認率會 造成致命之錯誤的情況，因此，亦仍多有著與其要求類似 性而更要求相同性之畫像檢索的需要。 如同上述一般’追求畫像檢索之確實性和時間之縮短 ，在目的上係相互矛盾而無法相容，但是，首先，針對在 畫像檢索中而確實地將相同畫像檢測出來一事作說明。 如同上述所說明一般，本發明係對確實性有所要求， 原則上，係爲將每1座標（位址）作爲比較檢測對象而進 行資訊之檢測者，以下，針對爲了實現此事而無法或缺之 資訊的種類以及其之解析度作說明。 畫像之資訊係存在有各種種類，但是，於此係大略地 將2種畫像資訊爲例來作說明。 第1，在將被作顯示之畫像的從圖框緩衝（圖形記憶 體）而來之資料作爲畫像資訊的情況時，通常，在彩色的 情況，係以R、G、B合計16bit至64bit之資料長度而持有資 訊。 雖然亦能夠將此R、G、B色訊號直接作利用，但是， 作爲效果性的畫像檢測之其中一例，通常係將電腦或者是 映像裝置之圖框緩衝的200萬像素左右作爲對象，而藉由 對R、G、B作各4bit的採取（16種組合中之3種）並作爲1 個的像素資料，來成爲就算是何種色彩之畫像均能夠以良 好精確度來將畫像檢測出來。 於此情況，此顏色之組合，係爲12bit、4096種組合， -16- 201137651 當畫面上之顏色有所不均的情況時’ 1個顏色存在於畫面 上之機率，係爲200萬/ 4096 4 48 8像素（位址）。 第2，作爲其中一例’亦可將JPEG (Joint Photographic Experts Group )或者是 MPEG ( Moving Picture Experts Group )等多數被作了壓縮之畫像資料的1個區塊（作爲其 中一例，例如8 X 8像素）作爲1個座標，並將該區塊之亮度 或者是色差訊號的DTC (離散餘弦變換）的DC (直流）成 分資料作爲該座標之該資料來直接作利用、或者是對於其 他資訊（例如向量資訊）作利用。 於此DCT之情況，由於係爲區塊單位，因此，相較於 像素，係能夠大幅度（作爲其中一例，例如1 / 6 4 )的減 少座標之數量（位址）。 當然的，不論是在何種情況中，均係以高解析度爲佳 ’但是，由於記憶體容量亦會變大，因此，只要從上述一 般之量子化資料的LSB側來選擇必要之bit數並作爲畫像資 訊資料即可。 以下，將上述所說明之像素作爲對象，並將由2 0 0萬 像素（位址）而1 2b i t、4 0 9 6種之R、G、B的組合資料所致 之衋像資料作爲畫像資訊，來對於對此進行檢測的情況之 例作說明。 [實施例1] 圖1 (取樣點之例），係爲對於當將從畫面之左上角 隅到右角隅爲止的丨、2、3、4.....η之η個像素111之資 -17- 201137651 料依據記憶體之位址1 0 3的1、2、3、4.....η之順序來作 配列記憶的情況時，成爲檢測基準之畫像（既知之資訊 1 〇 1 )的取樣點1 1 3作展示者❶ 檢測基準畫像A，係以尺寸較小之畫像作爲對象，並 爲在成爲檢測之基準的畫像（既知之資訊丨〇丨）的領域上 ’以座標y〇 ' x〇作爲中心，而上下左右等間隔地將合計25 個座標作爲取樣點1 1 3來作了自動配列的情況，於此情況 ’係爲在XY軸處均各將33像素111，亦即是合計1〇89像素 1 11作爲對象。 就算是將檢測基準之畫像的尺寸增大，亦完全沒有問 題。 圖中所示之1〜2 5的數字，係代表樣本之比較順序， 在本例中’係以中心作爲基點，並設定爲從距中心較遠之 取樣點1 1 3起朝向較近的取樣點u 3來依對角順序而作檢測 ’但是’配列或者是比較順序，係並不被限定於此，關於 此事之詳細內容，係於後再述。 檢測基準畫像B，係爲並非如同以上所說明—般之以 等間隔來作配列，而是與檢測基準之畫像相對應地而將i 〜22之取樣點1 1 3藉由手動來作了設定的情況之例，如此 這般之藉由手動來設定取樣點1 1 3並進行檢測的方法，係 爲對於人類之高特徵辨識能力作了利用者，在對於特徵性 之點或者是特徵性之範圍作指定並謀求與其他像素之差別 化一事上，係爲有效。 經由如此這般之爲了謀求畫像檢測之確實性而將1個 -18- 201137651 座標作爲檢測單位的檢測方法’而能夠將在每1座標處進 行取樣點設定一事，在多係將一定區域統籌性地作爲特徵 點的其他檢索、檢測方式中，係並無法實現’而亦身爲本 方式的特徵之一。 [實施例2] 圖2 (由取樣點所進行之資訊檢測實施例），係爲對 於將先前所說明之圖1的檢測基準畫像A作爲檢測之基準的 畫像並實施了畫像之檢測的情況作說明者。 係展示有：成爲檢測之基準的畫像（既知之資訊1 〇 1 )、和成爲檢測之對象的畫像（未知之資訊1 〇2 )。 在未知之資訊102的圖中所展示的群組A係爲雖然在1 次比較中爲一致但是在2次比較中卻成爲N G的情況，群組 B係爲在11次比較而成爲N G的情況，群組C係爲在2 2次比 較而成爲NG的情況，群組D係爲全部之取樣點1 1 3均爲一 致的情況，此些之檢測，係只要對於先前所說明之座標的 資料而在每一位址處作讀取並相互地作比較即可。 在此圖中，雖係爲藉由從群組A至群組E均係在位置上 完全被分離之畫像區域而作了說明者，然而，在畫像的情 況時，通常，身爲同一資料之座標，係爲相鄰接或者是集 中且若是解析度越低則此種傾向會越顯著，但是，只要對 於相互間之相對位置關係作正確的比較，便不會有問題。 於此情況，雖然在機率上，群組D之區域亦充分可能 會被判斷爲與成爲基準之畫面同一之畫像，但是，藉由除 -19- 201137651 了取樣點113之外更進而對於相互之畫像的全部之像素彼 此間的一致作再度的確認，係成爲能夠將類似之畫像排除 並保證其爲完全相同的畫像，此一方法，在成爲對象之畫 像的尺寸較小之情況的檢測時，係最爲適合。 藉由對於上述一般之檢證方法作活用，當如同群組E —般之取樣點Π 3部分性地集中並成爲不一致一般的情況 時，亦可將此一致之畫像的部分判斷爲在一部份之畫像處 被施加有變化之變形畫像，又，在如同群組C —般的情況 時，亦可判斷爲類似畫像，針對此些情況，係於後再述。 如同上述一般之先將某處之取樣點113的1座標作爲基 準，並進行之後的座標之資料間的相對位置之判定，再將 在相對位置中並不存在矛盾的座標作爲候補座標而殘留的 方法，係爲在圖案辨識等之資訊檢索中的常用手段，對於 此些之組合性的檢索有效率地作實施並將畫像（資訊）超 高速地檢測出來的方法與裝置，則係爲本發明之精髓所在 ，關於其詳細內容，係於後再述。 在以上所說明之畫像的檢測方法中，特別重要之事項 ，在取樣點113之選擇方法中，其中1個係在於一定之畫像 資料的範圍，另外1個則係在於畫像資料之變化的程度。 例如在對於不存在變化之黑畫像部分或者是白畫像部 分作了指定、或者是對於變化爲少之畫像作了指定、亦或 是對於例如僅存在有文字資訊之畫像等的特徵爲少之畫像 作了指定的情況時，該畫像之檢測係成爲困難。 以下’展示針對與畫像之檢測的有效性、信賴性存在 -20- 201137651 有深切關係之取樣點1 1 3所作的考慮。 若是將畫像上之1個座標作爲基準，則與此座標相鄰 接之座標的與基準座標成爲相同或者是近似之量子化資料 的機率（亦即是相關性）係變高，並隨著座標之遠離而使 得相關性變低，故而，係以如同圖1之樣本一般地每次均 從較遠之取樣點1 1 3來依序作確認，其之一致、不一致的 判斷變快並成爲有效率之檢測的機率爲較高。 故而’由完全地被作了分散化之複數的取樣點1 1 3全 體所致之同一座標群的存在機率，係成爲在此些之取樣點 113的資料之bit數上積算了取樣點113之數量的指數。 例如’在先前所說明之1座標爲R、G、B各4bit構成的 情況時’組合數係存在有2的1 2次方（4 K種），但是，若 是取樣點113爲10個場所，則係成爲2的120次方之組合數 ’在實際動作上，係成爲相當於無限大的機率組合，只要 彼此並非爲特徵爲少之單調的畫像，則若是此些之全部的 取樣點1 1 3均爲一致，則係可判斷爲同—畫像。 然而’由於亦會有以受限之狹窄範圍之畫像作爲對象 的情況’或者是例如文字等之黑白畫像亦會成爲對象，因 此’並無法採取上述一般之完全地被作了分散化的取樣點 113» 故而’係只要進行以下所說明之取樣點1丨3的辨識能 力評價’並採取例如發出警報或者是將畫像區域擴大亦或 是將取樣點1 1 3作追加等之適當的對策即可。 -21 - 201137651 [實施例3] 圖3 (取樣點之評價方法例），係爲圖1之檢測基準畫 像A的情況時之對於此取樣點1 1 3的辨識能力作評價之例， 並對於將座標1作爲檢測基準座標而將最後設爲座標25爲 止之合計25個的取樣點113，來將相鄰接之4個取樣點113 作爲1組，而出現有A〜P之合計1 6組的鄰接取樣點群。 作爲其中一例，在A群中，係包含有2、1〇、14、18之 4個取樣點113，在B群中，係包含有10、6、18、22之4個 取樣點U 3，以下亦同。 此時，A群〜P群之各別的4個取樣點1 13，在各別之群 中，不論是於亮度資訊或者是顏色資訊的何者之中，均在 座標之資料中存在有差異一事，由於係與特徵之大小、亦 即是與樣本特徵量之大小有所相關，因此，藉由從此4個 樣本中採取2個所成之組合的6個組合、亦即是2 -1 〇、2 - 1 4 、2-18、 10-14、 10-18、 14-18，來求取出其之資料的差分 量之絕對値，並取得此6個組合之合計和全體（16群）之 集計，而能夠設爲該群之特徵量的特定値。 當如同本例一般的使R、G、B作了複合之資料的情況 時，只要各別獨立地作評價即可。 又，當對於同一樣本數之情況的辨識能力作掌握的情 況時’係可將A群〜P群之各個的特徵量作合計並除以〗6群 ，而將所得到的平均特徵量作爲特徵量的大小之尺度（特 定値）來作利用。 當然的，若是此樣本之特徵量爲少，則由於會對於作 -22- 201137651 爲取樣點1 1 3之辨識能力造成影響，因此，當對於檢測之 基準畫像作指定並決定其之畫像區域時，只要以使其成爲 基準値以上之値一般地，來因應於必要而進行將取樣點 1 1 3之數量增加或者是將畫像之區域增加等的調整即可， 不用說，亦可能會有相反的情況，此取樣點1 1 3之評價方 法，在將畫像之檢測的檢索處理數（時間）設爲合理之檢 索處理數（時間）一事上，係爲重要。 以上內容，在藉由手動來決定取樣點113時，當然亦 爲有效。 本例，雖係爲將二維資訊作爲對象來對於辨識能力作 了評價之其中一例，但是，係能夠藉由以求取出從一維至 多維之相鄰接的樣本和其之資料的差分一事所致的想法， 來判定出特徵量。 進而，依存於成爲對象之資訊的種類，若是根據該資 訊之特徵來對樣本之基準獨自作定義並進行判定，則係成 爲能夠進行更確實之資訊的檢測。 至此爲止，係針對了對於在實現本發明時所不可或缺 之確實性作了追求的資訊之檢測，而以二維資訊之畫像爲 對象而作了說明，但是’此一想法之基本內容，對於其他 維度之資訊，亦爲共通。 於此，針對用以實現本發明之記憶在記憶體中的資訊 之配列作說明。 一維配列之資訊，係爲在記憶體位址上而被連續性地 作了記憶的資訊，二維之資訊，係爲如同圖1之1〜η的位 -23- 201137651 址1 03 —般之將個別的維度之最大座標數作爲配列基本條 件而被作表（tab 1 e )變換並作爲一維之記憶體位址配列而 被連續地作了記憶或者是可作記億之資訊，且爲使記憶體 之資料大小以及記憶體容量分別與各自之目的相互對應了 的記憶體構成。 當將3維甚至多維之資訊記憶在記憶體位址上的情況 時’亦同樣的，此些之資訊，係將個別之維度的最大座標 數作爲配列基本條件並被作了表變換之與上述相同的資訊 ，且記憶體構成亦爲相同。 故而，此資訊，若是被賦予有與各個的維度相對應之 座標資料，則係能夠將與根據身爲配列基本條件之各維度 的最大座標數而被作了賦予的座標資料相對應之位址特定 出來，且座標之相對位置或者是座標之範圍亦同樣的能夠 特定出來。 當然，在高維資訊的情況時，亦同樣的，當能夠直接 得知作爲對象之位址的情況時，係能夠直接以該位址或者 是相對位址來作指定。 以上，係爲爲了確認而再度作補足記載者，此位址配 列方法，係爲通常所進行之一般性的資訊配列方法，此一 般性之對於記億體的資訊之配列方法，由於係爲用以實現 本發明之基本資訊配列，因此，係極爲容易作利用。 之後，依據至此爲止之想法，來針對將本發明之具備 資訊過濾檢測功能的記憶體1 2 1適用在聯想記憶體中的情 況作說明。 -24- 201137651 聯想記憶體，係被利用在快取記億體（cache memory )或者是通訊資料處理等之在資訊處理裝置中而特別重要 且需要高速處理之記憶體中。 又，係作爲對於畫像資料等而以特別是由叢集手法所 致之用以檢測出類似畫像的最短距離檢索（類似度距離檢 索）作爲目的之檢索裝置而被作了廣泛的硏究。 接著，對於聯想記憶體之槪要作說明。 在高速之資料處理中所被作利用的聯想記憶體，係除 了通常的記憶體功能之外，亦能夠從外部而將作比較之資 料對於全記憶體而同時（並列）地作賦予，並將其之合格 的記憶體之位址讀出的在資訊之檢索上非常便利的裝置。 若是對於此並列處理舉出一個想像例，則例如在聚集 有大量的人之會場中準備座位（位址）並請人們坐下，且 設爲能夠讓這些人們自由地選擇喜歡的顏色之卡片（資料 )的情況時’例如針對持有紅色卡片（資料）的進行調査 時，在通常之記憶體的情況下，係有必要對於全部的人而 依照座位順序等來藉由逐次比較而作調査，相對於此，在 聯想記憶體的'丨η況下’由於例如係只要請拿著紅色卡片的 人一起將手舉起（並列比較）’再對於其之座位（位址） 作確認（輸出處理）即可，因此，係能夠進行極爲高速的 判定。 如此這般’聯想記億體，在對於大量的資訊作並列比 較’並從其中僅將必要之資料尋找出來的情況時，係成爲 便利的裝置。 -25- 201137651 聯想記憶體雖然具備有各種的優點，但是，作爲構成 上之弱點的其中之一，當進行由資料匯流排與位址匯流排 所致之資料讀寫的記憶體裝置之情況時，就算是能夠對於 從外部所賦予之比較資料（於此情況，係爲紅色）而同時 地進行其之合格判定，當合格之位址爲複數的情況時，係 並無法一次地將該位址作輸出。 爲了對此作解決，係只要在輸出處使其具備有優先度 (priority )功能，並對於合格之記憶體而依序將合格之 記憶體的位址輸出即可，但是，若是並未成爲充分地作了 過濾的位址數，則在讀出時亦會需要耗費時間。 在通常的情況中，係將此位址讀出，並對於被讀出了 的位址賦予下述之比較條件，之後，係根據此條件而進行 由逐次處理所致之過濾。 在先前之例子中的200萬像素（位址）而爲12bit、 4〇96種之組合資料的情況時，係有必要將平均48 8個的位 址作爲對象，並反覆進行之後的條件比較，通常，其中之 大半係爲對象外（不會殘留）之像素，而爲不必要的處理 ，在3次以後，亦爲相同。 故而，本發明之具備有資訊過濾檢測功能的記億體 121 ’係爲了將此種逐次處理完全地排除，而進而具備有 :在持有紅色卡片之將手舉起的人之中’使除了其之座位 的關係（位址之關係）爲一致的人（例如鄰居彼此或者是 前後左右持有紅色卡片的人）以外的人一齊將手放下的功 能’亦即是經由賦予資料與其之位址之關係、也就是以全 -26- 201137651 記憶體（座位的人）作爲對象並賦予比較條件（卡片之顏 色、座位之相對關係），來對於被記憶的資料之合格（卡 片之顔色）'和其之位址彼此間的位置關係（座位之相對 關係）之合格作判定，再過濾並檢測出在由此雙方之並列 合格與否判定結果所致的邏輯積（AND )演算中而合格了 的位址（鄰居彼此或者是在前後左右持有紅色卡片之座位 的人）’而將其作輸出的功能，亦即是，此記憶體，係爲 一種能夠進行雙重並列合格與否判定之記憶體，並進而實 現一種能夠進行各種之雙重並列邏輯演算的記憶體。 圖4 (資料以及位址之雙重並列邏輯演算的槪念）， 係爲以上所說明之記憶體的資料與其之位址的雙重並列合 格與否判定結果之邏輯積（AN D )演算的槪念。 亦即是，係爲用以將對各個的位址之資料的內容之比 較、和對各個的位址之比較，而分別並列（雙重並列）地 作合格與否判定，並根據該合格與否判定結果，來進行資 訊之過濾的邏輯積（AND)演算，更進而並列地來進行者 〇 此演算結果，不論是被藉由何種形態來作利用均無妨 0 如同上述一般，就算是1次的過濾，亦能夠得到極大 的過濾效果，但是，若是進而將此些之雙重並列邏輯演算 連續反覆地進行，則使理想之資訊過濾檢測成爲可能。 爲了實現以上之想法而將圖4之下段的位址比較之位 址合格與否判定電路作爲槪念來表現一事，雖係爲容易， -27- 201137651 但是，在通常之想法中，由於要如何地對位址作比較、以 及要與何者之位址作比較一事，係並非一定，因此，要實 際地將此槪念作邏輯電路化一事，係並不容易。 例如，雖係可考慮：將至此爲止所說明的作比較並殘 存的位址，作爲1次突破位址，並將此作爲基準，來與各 位址進行比較之方法，但是，在此方法中，亦同樣的如同 前面所說明一般，若是假設1次突破位址係存在有48 8個位 址，則由於係成爲必須要構成由全部之位址與此48 8個位 址之組合所致的組合並列位址比較電路，因此，係成爲極 爲龐大之規模的構成。 若是小規模的記憶體位址數，則雖然亦能夠藉由上述 之構成來實現位址比較，但是，在本實施例中，係爲了就 算是大規模的記億體亦能夠將圖4之邏輯構成儘可能地藉 由簡單的電路構成來實現，而對於下述之情況作注目：亦 即是，藉由將1次突破位址定義爲每次之用以作比較的基 準原點之位址，之後之每一次的作比較之位址（座標）， 此各個的1次突破位址與相對位址（座標）會在每一次之 比較中均成爲相對性的同一位置（位址），也就是說，係 藉由對於1次突破位址和之後每次作比較的相互之位址彼 此間的相對關係作比較’來求取出此邏輯電路的最適解。 具體而言，係只要設爲下述之構成即可：亦即是，對 於在先前所說明之圖1中所示之各個的取樣點113，將1次 突破位址作爲基準原點座標，並在作比較之位址丨〇 3中設 定一定之相對性的偏移或者是範圍，而對於各個的！次突 -28- 201137651 破位址之目標的位址1 〇3之資料是否合格於此~事作確認 ，若是合格，則將此設爲突破位址。 進而，經由在各個的位址103處設置對於突破之次數 作記錄的計數器，並設爲能夠在身爲基準原點之1次突破 位址處將突破次數作累積而作正數（count up)，能夠藉 由將1次突破位址中之最多突破次數（N次）的記憶體之位 址判定爲N次突破位址，來使連續過濾之邏輯電路成爲可 能，就算是極爲簡單之邏輯電路的構成，亦能形成爲滿足 當初之所有目的的構成，而能夠實現一種對於資訊處理之 長年的課題之其中一者作了克服的裝置。 雖係於後再述，但是，圖4之雙重並列邏輯演算，係 並非僅被限定於邏輯積（AND )演算，又，此演算結果之 利用方式，亦爲自由。 [實施例4] 圖5 (具備有資訊過濾檢測功能的記憶體例），係爲 將與以上之內容相關連的本發明之記憶體！ 2〗之功能槪要 以聯想記憶體作爲基礎來作了展示者，資料處理之時序等 的細部構成’係被作省略，並僅對於與本發明相關之處的 槪念作說明。 在此一具備有資訊過濾檢測功能的記億體！ 2 1 (之後 ’係亦記載爲本發明之記憶體）處，係被連接有位址匯流 排1 22、資料匯流排I 23，並成爲能夠與外部進行資料之授 受的構成。 -29- 201137651 故而’記憶體1〜n之記億體1 32，係藉由位址匯流排 122之位址解碼器131而對於位址丨〜η作選擇，並能夠從資 料匯流排1 2 3來進行資料之寫入、讀出^ 輸入資料125 ’係爲對於本發明之記憶體121賦予用以 進行資訊檢測之資料者，身爲第1資料之記憶體比較資料 1 26 ’係爲從外部而來之用以進行記憶體之資料比較的資 料’將此輸入資料125與記憶體1〜η之記憶體132間的資料 之合格與否的比較，藉由資料比較電路1 3 3來作判定，當 合格的情況時，係將該結果作爲位址置換前合格輸出1 4 1 而作輸出。 針對身爲第2輸入資料125之位址比較資料127、以及 位址置換電路134，係於後再述。 突破次數計數器135，係爲藉由位址置換合格輸出142 而將資料比較電路1 3 3之合格次數作爲突破次數來作記憶 加算的計數器，此突破次數計數器135，係具備有與對於 資訊彼此間之比較次數作計數的比較次數計數器1 29之比 較次數訊號143間的一致輸出功能，其之輸出，係被與OR 閘136和條件閘（inhibit gate) 137作連接，並爲藉由從較 小的位址來依序作級聯連接1 44之訊號，而使對於在突破 次數爲N次的突破次數計數器135之中爲最小的位址的計數 器爲優先而進行僅作1個位址的輸出之輸出優先（priority )處理。另外，1 2 8係爲重置訊號。 突破位址輸出處理電路138’係藉由進行將優先輸出 之位址送至輸出匯流排1 24上之處理、和將已結束輸出處 -30- 201137651 理之位址的突破次數計數器135作清除之處理，而成爲： 在之後’若是存在有其他的N次突破之突破次數計數器135 ，則能夠將其之位址作爲下一個的優先輸出，並依序將N 次突破之位址經由輸出匯流排1 2 4來送出至外部的構成。 本例之專用匯流排輸出的專用輸出形態，係僅爲其中 —例’亦可直接將輸出結果送至資料匯流排1 2 3處。 故而’若依據此構成，則突破次數最多（N次）的突 破次數計數器135之位址（座標）係成爲Winner(N次突破 位址）’並依其位址由小而大的順序起將該位址作輸出。 [實施例5 ] 圖6 (位址置換電路之第1例），係爲極爲簡單之邏輯 電路構成’並爲對於身爲用以實現本發明之手段的位址置 換電路134之基本槪念作展示者。 位址置換電路134，係被設置在資料比較電路133與突 破次數計數器1 3 5之中間，此位址置換電路1 3 4，係爲爲了 在每個樣本的比較時，在目的之1次突破位址，將突破之 輸出作爲雙重並列邏輯積演算結果來作累積加算所被設置 者’在本例之情況中，係構成爲：能夠將身爲先前所說明 之輸入資料125的第2資料之位址比較資料127之經由相對 位址比較資料而將位址置換前合格輸出1 4 1、圖6之i、j、k 變換爲χγ軸座標資料，並將變換後的合格輸出作相對位 址之量的移位（shift)，再作爲位址置換後合格輸出142 來在該當之位址的突破次數計數器1 3 5 ( 1次突破位址）處 -31 - 201137651 將合格輸出作爲突破輸出而作輸入。 亦即是，位址置換後合格輸出1 42，係當1次突破位址 之位址的相對位址條件爲合格的情況時，作爲突破輸出而 被輸入至1次突破位址處。 當然，亦可並非爲座標資料，而是將相對位址比較資 料直接藉由相對位址來作指定並作相對位址之量的移位。 以上所說明之第1、第2資料的輸入，要從資料匯流排 123來作賦予或者是從專用輸入來作賦予一事，係爲自由 〇 針對將先前所說明之畫像的像素資料或者是與此相當 之資訊資料記憶在上述構成之本發明的記憶體1 2 1中，並 根據此未知之資訊1 02來將畫像檢測出來的情況之例作說 明。 在先前所說明之圖5的記憶體1〜η之記憶體132中，畫 像的像素資料係被寫入至與各別之座標相對應的位址處， 比較次數計數器129以及全部的突破次數計數器135，係全 部被作了清除而成爲〇，之後，比較次數計數器1 29係在每 一次的比較中而被作正數（count up)。 首先，作爲1次比較，將樣本1之像素資料作爲輸入資 料125而賦予至記億體比較資料126處，並列進行全部記憶 體之合格判定，將資料比較電路133之位址置換前合格輸 出141作爲1次合格輸出來作輸出，此1次合格輸出，係並 未被作位址置換，並作爲位址置換後合格輸出142來直接 地加到突破次數計數器135之輸入處，而將突破了的位址 -32- 201137651 之計數器的値設爲1 ’此係爲1次突破位址，如同上述一般 ，在1次比較中，係並不需要第2資料。 如同先前所說明一般，平均性之1次突破位址的出現 個數係爲488 (圖6之i、j、k) ’此出現個數，由於係爲假 設上的數字，因此’就算是較多或者是較少也無妨，以下 ，亦爲相同。 於圖2之情況’此時，突破次數計數器1 3 5之値成爲1 的記憶體位址，係爲群組A之1、群組B之1、群組c之1、 群組D之1以及群組E之1的5個場所之座標成爲winner的候 補（1次突破位址），此係爲在後述之說明中而作爲重點 的位址（座標）。 接著’作爲2次比較’藉由將樣本2之像素資料指定爲 記億體比較資料126，而又將其他之位置重新作爲2次合格 輸出來平均並選擇48 8個。 進而，藉由將樣本1與樣本2之位址的差分指定爲位址 比較資料1 2 7的相對位址比較資料，在新選擇了的4 8 8個的 位址中’將與先前所說明之1次突破位址間的相對關係爲 合格者、亦即是群組之關係爲成立的位址，藉由圖6中所 示之位置置換電路1 3 4來對相當於此差分之位址（座標） 作移位變換’並在作了移位變換之相對位置的突破次數計 數器1 3 5 ( 1次突破位址）處，將位址置換後合格輸出1 42 作爲突破輸出而作加算輸入。 亦即是，原本係爲圖2之從群組A〜E之2的座標（位址 )之突破次數計數器135爲被作正數（count up)，但是， -33- 201137651 在樣本1之判定中所生存之候補座標的突破次數計數器135 (1次突破位址）處，係以能夠繼續地作正數（count up ) 的方式而被施加有相對位址之偏移，並作爲2次突破輸出 而對於1次突破位址賦予突破輸入。 於先前所說明之圖2的情況，此時，突破次數計數器 1 3 5之値成爲2的1次突破位址，係爲將群組B之1、群組C之 1、群組D之1以及群組E之1的4個場所作爲候補而作了維持 之座標，群組A之1係並未被作正數（count up )，而被從 候補中剔除。 以上之內容，係與「將作爲對象之記憶體的位址位置 (相當於2次樣本之位址位置）以1次突破位址作爲基準來 判定其是否存在於目的之位置（相對位址比較資料）處， 並和藉由2次之資料比較所致的合格位址並列地進行邏輯 積（AND )演算而將其結果作爲突破輸出來輸入至1次突 破位址之突破次數計數器135中」一事爲等價。 依序同樣地將記憶體1 32之資料比較和以1次比較之樣 本1作爲基準的與其他樣本間之相對位址，作爲一對之輸 入資料125而作讀入，並在各個的位址群組內而將突破的 記憶體集中在1次突破位址中而作正數，藉由此，而能夠 連續地進行1次突破位址（Winner候補）之過濾。 故而，在圖2中，一直突破至了最後的樣本25 (N爲25 )之1次突破位址，係僅有群組D之1的座標（25次Winner )，此座標（位址）的突破次數計數器135之値，係成爲 25，此突破次數計數器135，係與比較次數計數器129之比 -34- 201137651 較次數訊號143相一致，該輸出，係被輸入至之後的〇R閘 136以及條件閘137處。 圖7 (位址置換之第1示意例），係爲將至此爲止所說 明之「資料的合格」和「位址之相對關係之合格」的雙方 之合格、亦即是由雙重並列邏輯積演算所致的突破之內容 作爲示意圖來作說明者》 如圖7中所示一般，在畫面之座標中，係藉由最初之1 次比較而展示有從A〜F之總計6個的1次突破位址。 此位址置換，係爲將全部位址作爲對象而相對性地進 行者，但是，1次突破位址A〜F，係如同在成爲比較之對 象的各個位置群組內，藉由望遠鏡來注視接著所被作比較 之2次比較的相對座標位置，並當其係爲2次合格位址之合 格輸出（在本圖中，係以黑圓來作標示）的情況時，則將 此作爲突破輸出而奪取，正如同進行置換動作一般。 3次比較，亦同樣的，係藉由望遠鏡來注視被作比較 之相對座標位置，並將3次合格位址之合格輸出（在本圖 中，係以黑三角來作標示）作爲突破輸出而作奪取，之後 ，亦進行同樣的置換。 於本例之情況中，E的計數器係被更新爲2，進而，B 的計數器係被更新爲3。 圖8 (位址置換之第2示意例），係爲將以上之圖7中 所示的A、B之2個1次突破位址之座標性的示意作爲位址性 之示意來作說明者。 如圖8中所示一般，1次突破位址A、B，其示意係如同 -35- 201137651 藉由望遠鏡來對於從樣本2起直到樣本25爲止的資料比較 電路作注視，並當在作了注視的資料比較電路中存在有合 格的情況時，將其作爲突破輸出而作奪取》 當然的，望遠鏡之切換，係藉由身爲用以對於位址之 相對位置關係作比較的資料之相對位址比較資料，而在每 —次中被作設定。 實際上，與1次突破位址A、B之位址間並不存在有相 對關係的由2次、3次、N次比較所致的合格輸出，亦在被 相對性地作了移位之位址處而被作正數，但是，若是樣本 並非爲合適且具有意義者，則係會在每次零星分散，合格 輸出係不會有集中在特定位址處的情況。 此係由於「成爲樣本之畫像（資訊）與未知之畫像（ 資訊）的特定部分係爲相同」的特別關係（圖案）並不成 立之故。 在示意上，1次突破位址，係恆常地保持有計數之優 位性（最初係爲1)，且爲進而具備有代表與1次突破位址 之相對位址附加有關連的樣本之位址的群組並將合格輸出 作爲突破輸出來作收集的權利之如同支配者一般的地位。 圖9(位址置換之第3示意例），係爲對於在實際的2 維配列位址中之位址置換之例作展示者。 表A、B，係爲對於位址置換前之位址（座標）1〜100 而作展示者，24、50、67、72之4個位址（座標）係成爲1 次突破位址。 表A，係爲將2次比較位址以相對位址-22之位址作爲 -36- 201137651 資料比較位址的情況’此時，7 2之位址’其之目標的座標 ，係爲對象外。 表B，係爲將3次比較位址以相對位址+ 3 1之位址作爲 資料比較位址的情況’此時’ 50以及72之位址’其之目標 的座標，係爲對象外。 表C，係爲將表A作了 - 2 2之位址移位者’ 2 4、5 0、6 7 之1次突破位址，係能夠正常地得到各別之目標的資料之 合格與否結果，且若是爲合格結果，則係能夠分別作爲突 破輸出而作正數（置換計數）。 表D，係爲將表B作了 + 31之位址移位者，24、67之1 次突破位址，係能夠正常地得到各別之目標的資料之合格 與否結果，且若是爲合格結果，則係能夠分別作爲突破輸 出而作正數（置換計數）。 將以上操作反覆進行特定次數，比較對象之目標的位 址之座標位置爲正常的I次突破位址，係能夠一直生存至 最後。 在以上之藉由各種例子所作了說明的至此爲止之內容 中，係與「藉由進行位址置換（位址的替換），來將〗次 突破位址作爲基準’之後’對於是否存在有與在每一次所 被比較之取樣點1 1 3相同之資料（資料之合格）並且進而 亦存在於目的之位置（相對位址比較資料）的兩者，而連 續性地且雙重並列性地來進行合格與否判定，並進而並列 性地進行邏輯積（AND )演算，而將其結果於每—次而輸 出至1次突破位址之突破次數計數器1 3 5中」一事相等價， -37- 201137651 位址置換電路1 34，係爲使圖4之下段的並列位址合格與否 判定與並列邏輯積演算作了 一體化，而成爲以一者而具備 2種功能之極佳效率的邏輯電路（亦即是等價雙重並列邏 輯積（AND)演算手段）。 最終結果，係只要將此身爲比較次數之N次突破位址 ，以突破位址輸出處理電路1 3 8以及輸出匯流排1 24讀出， 則係能夠將包含有N次突破位址之資訊的群組之位址特定 出來，亦即是作了圖案辨識。 藉由將比較次數計數器1 29作爲可預置計數器來對於 比較次數訊號1 43作指定，係亦能夠將任意計數値的計數 器（N次突破位址）之位址或者是其之途中經過讀取出來 〇 位址置換電路134之其中一例，係如同圖6中所示一般 ，準備位址變換用之暫存器，並將藉由資料比較電路133 所得之位址置換前合格輸出1 4 1，藉由相對位址比較資料 之座標資料來相對性地作移動置換，並設爲位址置換後合 格輸出1 42，此一暫存器操作，由於係爲全部之位址的相 對移位，因此，就算是經由以加減算演算所進行的資料移 位手段、或者是經由最爲簡單的具備位址之量的資料長度 之移位暫存器，亦能夠容易地實現。 以上之使用暫存器的位址置換方法，係爲用以進行說 明之其中一例，亦可藉由像是直接利用位址解碼器來進行 置換（位址置換）等等之其他方法來實施之》 同樣的，位址置換電路134以及突破次數計數器135， -38- 201137651 雖然係爲在實現本發明之資訊過濾檢測時所不可或缺的手 段，但是，並不被限定於此構成，亦可藉由其他方法，來 在每一位址處個別實施。 對於資料與其之位址的相對關係整批地進行合格與否 判定之方法，由於係與「將圖4中所示之1〜η的全部記憶 體1 3 2之資料比較條件的合格和全部位址比較條件之合格 的雙方雙重並列地進行合格與否判定並進而對於此而並列 地進行了邏輯積演算」一事爲等價，因此，原理上，係爲 使將個別位址作爲對象之馮諾依曼型資訊處理的位址逐次 處理成爲不必要者。 故而，當將至此爲止所說明了的200萬像素之未知的 畫像設爲解析度12bit、資料群組數4096群組的情況時，從 未知之畫像中而進行同一畫像之檢測時所需要的資料之比 較次數’通常係在2或3次處而收斂，就算是最大亦可藉由 樣本數（於本例中係爲25次）之資料比較次數來確實地尋 找出目的之畫像。 進而’此方式，由於係如圖7中所示一般，爲由以1次 突破位址A〜F作爲基準原點的輸入資料所致的全座標圖案 匹配之反覆進行（每次將全部記憶體作爲對象而比較）， 因此’亦能夠實施像是部分性的而欠缺了 —部份之畫像一 般的情況時之近似的畫像（以下，稱爲近似畫像）之檢測 〇 例如，在圖2的情況時’於2 5次之比較結束後，群組A 之1的座標之計數器的値係爲1，群組B之1的座標之計數器 -39- 201137651 的値係爲1 〇，群組c之1的座標之計數器的値係爲2 1，群組 D之1的座標之計數器的値係爲25，群組E之1的座標之計數 器的値係爲22。 亦即是，計數器値爲高之座標的位址，其之身爲部分 畫像有所欠缺的畫像或者是近似之畫像的可能性係爲高， 於此情況，就算是1次突破位址以外亦無妨。 只要將計數器値成爲一定値以上（例如20次）之座標 預先讀出，並在特定（於此情況，係爲25次）的比較結束 後，依據需要而對於週邊之座標作詳細的判定即可。 故而，此方式，係並不僅是在以高速而將同一畫像檢 測出來的情況時爲有效，在檢測出依據一定之定義的近似 畫像時，亦爲有效。 又，藉由對於此一位址置換（位址的替換）之想法作 更進一步的發展，亦能夠將使畫像作了擴大縮小或者是旋 轉（以下，稱爲變形畫像）的畫像藉由最少的檢測次數而 檢測出來。 [實施例6] 圖1 〇 (變形畫像之檢測的槪念），係爲對於作比較之 畫像爲被作了擴大縮小或者是旋轉、乃至於依存於情況而 在資料上施加有變化之變形畫像作爲前提而進行檢測的情 況時之有效的方法作說明者。 圖10，係爲在1次樣本之比較爲合格者的未知之畫像 的1次突破位址上，重疊了既知之畫像的取樣點113者。 -40- 201137651 於本例中，係對於存在著以座標1之基準原 心而使未知之畫像的尺寸在X、γ軸上均作了 2倍 面，係成爲4倍）之擴大的可能性的情況作展示。 當所找出之變形畫像係存在於此之中的情況 形畫像相對應之樣本的2〜2 5之全部的座標，應 於圖中所示的圓之內部，故而，只要將包含有圓 圍作爲此座標1之畫像檢測範圍即可。 故而，係將至此爲止之位址（座標）移位變 擴大，藉由判定出在以座標1之基準原點所指定 圍中是否存在著具備有與樣本相同之資料値的座 於此情況，係無關於個數，而僅是單純地判定有 即是判定出與所指定之樣本數（於本例中，係爲 樣本相當之座標是否存在於此範圍中，而成爲能 畫像檢測出來。 於此情況，亦同樣的，藉由判定出樣本之特 樣本之數量與辨識能力設爲一定之基準而將對象 爲一定之範圍內一事，其之確實性係提高。 於此情況’亦同樣的，只要能夠實現每次將 址雙方之合格（突破）的輸出集中在目的之1次 處並作記憶加算之手段（於本例中，係爲使判定 積在1的座標之突破次數計數器135處的方法）， 對於此種變形畫像，亦能夠藉由將樣本數設爲最 之資料比較次數來進行畫像之檢測。 點作爲中 (作爲畫 時，與變 該會存在 之座標範 換的槪念 之座標範 標一事（ 無）、亦 25個）的 夠將變形 徵量並將 之範圍設 資料與位 突破位址 之結果累 則就算是 大且最少 -41 - 201137651 [實施例7] 圖22 (位址置換手段之第2例），係爲爲了實現上述 之想法，而將圖6中所說明之位址變換從1對1之位址移位 變換而發展成作爲位址之範圍來掌握，並將此從外部而作 爲位址範圍比較資料來輸入至位址比較資料127處，藉由 此，來將此位址之範圍內而爲合格之i、j、k的位址置換前 合格輸出1 4 1，作爲該當於比較條件之位址範圍而作讀入 ，於此情況，亦同樣的，係構成爲將位址置換後合格輸出 142輸入至各個的位址群組內之1次突破位址的突破次數計 數器135中。 在示意上，係如同將之前在圖7中所說明之望遠鏡更 改爲拋物線型之天文望遠鏡，並在1次突破位址處奪取突 破輸出。 例如，在將以至此爲止所說明之圖1中所示的檢測基 準畫像A之33 x3 3 4 1000像素左右作爲檢測對象畫像範圍的 畫像之1座標作爲基準的情況時，只要將該1座標作爲中心 ，並將XY軸7000像素強之座標範圍設爲比較範圍之座標 ，則就算是在被擴大成了 2倍之伴隨有旋轉的變形畫像的 情況時，亦能夠以將樣本數設爲最大且最少之資料比較次 數來進行畫像之檢測。 在被進行了座標變換或者是畫像被作了縮小之畫像等 的情況時，由於亦會有與樣本座標相對應之座標欠缺的情 形，因此，係只要適當地設置樣本合格次數之基準即可。 此方法，若是與圖6中所說明之相對座標的完全一致 -42- 201137651 方法作比較，則確實性係會降低，但是’藉由對於樣本之 辨識能力或者是將其數量作適當的設定’係能夠進行極爲 高速之畫像的檢測。 進而，若是使記憶體132之資料比較電路133，從資料 一致之合格與否的比較而作爲具備有亮度或者是顏色的準 位之範圍的大小比較來進行合格與否之判定，則不僅是變 形畫像，依存於其之定義方法，亦能夠檢測出類似的畫像 〇 如同上述一般，若是除了由完全一致所致之合格與否 判定以外，亦利用大小比較、範圍比較、記憶體bit個別比 較、乃至於能夠進行由Don’t Care所致的比較之3値記憶體 等，而進行合格與否判定，則能夠進行更有效之資訊檢測 〇 通常，施加了畫像之尺寸變化或者是旋轉的畫像之檢 測，係需要座標變換等之極爲龐大的檢索處理次數，但是 ，若依據此方式，則僅需要進行樣本數之量的比較次數， 便成爲能夠將目的之變形畫像（亦包含近似畫像）檢測出 來。 在多數之情況中，幾乎均爲如同上述—般’只要能夠 將變形畫像或者是類似畫像之中心位置或者是重心位置等 檢測出來即可，但是，就算是在有必要檢測出畫像之擴大 縮小或者是旋轉角度的情況時，亦只要追加數次之資料比 較即能夠作對應。 在存在有上述一般之需要的情況時，一旦將畫像所存 -43- 201137651 在之範圍檢測出來，之後只要如同圖10中所示一般，對於 2、4、3、5之4個對角座標係存在於何處一事，如同分割 檢測範圍A之4分割、分割檢測範圍B之1 6分割一般的作分 割，來對於範圍作限定並檢測出來即可，在4分割的情況 時，只要進行16次，在16分割的情況時，只要進行64次， 而合計進行最大80次之資料比較，便能夠對於該畫像的變 形之態樣作掌握。 通常，當檢測出此種無法對於變形之程度作推測之變 形畫像的情況時，係有必要對於所考慮之畫像的變形情況 作推測並作多數之座標變換，而進行圖案匹配，相較於此 種變形畫像的檢測，本發明係能夠進行無可比擬之高速的 圖案匹配。 若是對於分割之範圍作細分化，則亦能夠進行更加正 確的檢測。 以上’雖係爲其中一例，但是，藉由如此這般地追加 最小限度之資料比較的次數，亦能夠進行複雜之畫像的檢 測。 在本例中，係對於以畫像有被作了擴大縮小乃至於旋 轉的可能性一事作爲前提而將全取樣點1 1 3作爲對象地作 了大範圍之範圍設定的情況之例來進行說明，但是，亦可 對於各個的取樣點1 1 3而個別地指定一定之範圍並進行檢 測’此方法’在對於資料與其之位址（位置）間之不確定 性作補完的目的上，當同一或者是類似之資料連續地存在 的情況時’係具備有重要的意義，若是根據取樣點113之 -44 - 201137651 位置或者是其之資料來進行類似畫像之定義，則係成爲能 夠廣泛地檢測出同一畫像、近似畫像、變形畫像、類似畫 像。 本發明之記億體121，除了至此爲止所說明了的座標 移位方式（使位址與作比較之相對位址相一致）和座標範 圍方式（使位址存在於作比較之座標的範圍內）之2個檢 測方法以外，也能夠對於例如作比較之座標的範圍外等而 進行位址置換之應用，且不論何者均係能夠僅藉由位址置 換電路134之位址比較資料127的資料設定而實現之，因此 ，係能夠將此些設爲一體化之構成，藉由將此些畫像檢測 方法作組合，係能夠進行更多樣化之畫像檢測。 在本例中，係爲了將說明簡單化，而對於將R、G、B 之顏色資料統合爲1個位址之資料的方法來作了說明，但 是’就算是使R、G、B之各個的位址相互獨立並作比較之 方法，亦能夠容易地實現之。 本發明之記憶體1 2 1，由於係爲可在能夠並列地對於 記憶體資料作比較之基本構造的記憶體、例如聯想記憶體 等之中，而藉由用以進行位址之置換（位址置換）的手段 、和將合格次數作記億之計數器、以及一般性之優先編碼 器來構成的極爲簡單之構造，因此，大容量化係亦爲容易 〇 又，不用說，本發明之記憶體1 21，係爲一種裝置， 其係將根據資料與其資料之位址間的相互關係之組合問題 的探索（比較）次數作根本性的解決，只要是適當地被作 -45 - 201137651 了選擇的樣本，便能夠保證有以樣本數作爲上限之最少的 資料比較次數，並且，亦能夠應用在由被作了叢集的類似 特徵之座標相關所致的畫像之檢測或者是其他之各種的資 訊檢測中。 在至此爲止的說明中，雖係針對對於資訊作反覆過濾 的情況而作了說明，但是，不用說，亦可進行將過濾設爲 僅進行1次比較、2次比較之單發性的資訊檢測。 如同上述一般，若是能夠使用具備有可自我進行資訊 檢測之智能性知識之本發明的記憶體1 2 1，則在資訊檢索 時，由於CPU或是GPU係僅需要賦予輸入資料並將其結果 作讀取即可，因此，係能夠將負擔大幅度的減輕。 又，由於係能夠進行極爲高速之資訊檢測，因此，當 記憶體之容量不足的情況時，亦可對於資訊作分割並實行 資訊檢測。 本發明之記億體1 2 1的每次之資料比較處理時間，由 於就算是在保守估計的處理時間爲假設每次平均1//秒的 情況時’不論是以何種尺寸之畫像作爲對象，亦能夠在數 V秒乃至數百μ秒的時間內來確實地將目的之資訊檢測出 來，因此，就算是在以動態畫像作爲對象之1格（圖框） 上的資訊檢測出來的情況、或者是欲檢測出來之既知的資 訊1 〇 1爲連續大量存在的情況時，亦能夠廣泛地作利用。 當然，藉由與對於本發明之記憶體121的記億體132作 存取並進行逐次處理之通常的CPU之間作倂合使用（併用 )’係成爲能夠進行更爲高度之資訊檢測。 -46- 201137651 以上’雖然以畫像資訊作爲中心地而對於本發明之槪 要作了說明’但是’此資訊檢測之方法和本發明之記憶體 1 2 1 ’在對於1維資訊（聲音等）或者是作爲多維空間而作 了配列的資訊之檢測中，亦爲有效。 [實施例8] 圖1 2 (位址置換手段之第3例）’係爲使得對於將圖6 之2維資訊擴大爲X、Y、Z之3軸並配列在3維空間中的資 訊之同一配列或者是類似配列檢測出來一事成爲可能的例 子，且當然亦能夠設爲N維空間，關於資訊檢測之具體例 ，係於後再述。 至此爲止所說明之具備資訊過濾檢測功能的記億體 1 2 1 ’係能夠以全部的記億體爲對象而對於突破次數作計 數，並藉由該突破之次數來將同一資訊以及類似資訊連續 性地檢測出來，但是’在記憶體數量爲大規模的情況、或 者是爲了將電路構成更加簡單化的目的上，亦可將突破次 數計數器135、OR閘136、條件閘1 37之電路數作削減。 [實施例9] 圖1 3 (資訊過濾檢測電路之削減例）之記憶體1 2 1 a， 係爲根據圖4、圖5中所示之本發明的記憶體121之基本構 想而對於電路數作了削減者，在如同至此爲止所說明一般 地被作利用之突破次數計數器1 3 5之後的電路之數量，於 通常的情況時，係爲在1次比較中所出現之1次突破位址的 -47- 201137651 數量（在至此爲止的說明中，係爲200萬位址而解析度 4096種，平均48 8位址），對於此事作注目，而將與此相 匹配的數量、例如突破次數計數器1 3 5以後之電路數，削 減爲記憶體1 3 2之位址的數量之例如千分之一或者是兩千 分之一，在圖中，係削減至A〜X之輸出。 於此情況，係只要將計數器，設爲能夠將在圖7中所 說明之各個的位址群組之1次突破位址作記憶並且將此位 址讀出之構成的群組別突破計數器1 5 8，並設爲能夠在此 計數器1 5 8之各個處而分別將1次突破位址讀出的構成即可 〇 在設爲此種構成的情況時，係亦可在位址置換電路 134中搭載簡單的位址演算處理器等來進行位址置換，藉 由如此這般地利用演算處理器來將位址置換（位址的替換 )之自由度提升一事，係亦能夠期待有更加多樣化之手法 的資訊過濾。 假設當1次突破位址之數量爲多並造成溢位的情況時 ，係只要設爲發出警報並對於1次比較之樣本作變更等即 可。 又，作爲對於電路數作削減的方法，係亦可設爲將雙 重並列邏輯演算切換至記憶體之每一記憶庫並作實施等的 電路構成。 能夠以全部之記憶體作爲對象並對於合格次數作計數 的方式，對於1次突破位址之出現數量係並沒有限制，而 爲理想，但是，就算是如此這般地對於資訊過濾功能的電 -48- 201137651 路構成作了簡單化之具備資訊過濾檢測功能的記憶體121 a ，亦能夠藉由至此爲止所說明之資訊檢測的方法來將目的 之資訊確實地檢測出來。 藉由設爲上述一般之構成，能夠將關於本實施例之記 憶體1 2 1 a的記憶體之位址數或者是其之bit數（亦即是記億 體容量）的自由度增加，又，亦可如同前述一般地設爲3 値記憶體。 圖1 4 (作了多重化之資料以及位址之二重並列邏輯演 算的槪念例），係爲將圖4所說明之雙重並列邏輯演算作 了多重化之例。 如同圖1 4中所示一般，本實施例之記憶體1 2 1 b ’係將 用以與記憶體之資料作比較的比較資料與用以與位址作比 較之資料、和雙重並列合格與否判定電路、以及雙重並列 邏輯積演算電路，分別各具備有2組，並成爲對於邏輯積 演算結果而更進而進行邏輯和（OR)演算並作輸出之構成 〇 此種構成，亦可將在圖5中所示之記億體1 2 1的電路構 成作應用。 藉由設爲此種構成，係能夠將2個圖案同時地檢測出 來。 本例，係爲多重化之其中一例，而亦可設爲2組以外 之多數的組合，且亦可將演算設爲邏輯積（AND )或者是 邏輯和（OR )以外的排他性邏輯或者是其他之任意的邏輯 演算。 -49- 201137651 藉由因應於所檢測出之資訊的種類或者是目的來如此 這般地進行多重化，並使用各種之邏輯演算的雙重並列邏 輯演算，係成爲能夠進行更加高度之資訊檢測。 若是將至此爲止所說明之圖4、圖14等作統合，而對 於本發明之記憶體1 2 1、1 2 1 a、1 2 1 b之構成作總括，則係 成爲： 一種具備資訊過濾檢測功能的記憶體，該記憶體，係 爲將資訊記憶在每一記憶體位址中並且能夠將該資訊作讀 出之記憶體，其特徵爲，係具備有以下（1 )至（4 )： (1 )用以將從外部所賦予並用以和被記憶在此記憶 體中之資料並列地作比較之第1比較資料、和用以與此記 憶體之位址的位址彼此間作並列比較之第2比較資料，此 些之各比較資料作輸入之輸入手段；和 (2 )藉由第1比較資料而和被記憶在此記憶體中之資 料作並列比較並對於合格與否作判定之手段；和 (3) 藉由第2比較資料而和此記憶體之位址彼此作並 列比較並對於合格與否作判定之手段；和 (4) 將以上（2) 、（3)雙方之合格與否的判定結 果在每一位址處而並列地作邏輯演算之資料與位址的各合 格與否結果之邏輯演算手段。 進而，作爲用以將此記憶體之資訊過濾設爲簡單之電 路構成的其中一例，係成爲一種具備資訊過濾檢測功能的 記憶體，其中，前述具備資訊過濾檢測功能的記憶體之前 述邏輯演算，係爲邏輯積（AND )演算，並且，係具備有 -50- 201137651 以下（1 )至（3 ): (1 )在初次之資訊檢測時，藉由前述第1比較資料來 和記憶體之資料並列地進行合格與否之判定，並將合格了 的至少1個以上之記憶體位址作爲1次突破位址而作記憶之 手段；和 (2 )在下一次以後之資訊檢測時，藉由新的前述第1 比較資料來和記億體之資料並列地進行合格與否之判定， 並將合格了的記憶體位址，藉由由前述第2比較資料所進 行之位址置換手段來作了置換之位址；和 (3)將突破了上述（1) 、 （2)之位址的邏輯積（ AND)演算之位址作輸出之手段。 進而，如同圖5、圖13等之中所示一般，爲了使此記 憶體連續地反覆進行雙重並列邏輯演算，係爲一種具備資 訊過濾檢測功能之進行邏輯積（A N D )演算的記憶體，其 中’前述具備資訊過濾檢測功能的記憶體，係具備有以下 (1 )至（4)： (1 )對於各記憶體位址而個別地對於由反覆作了特 定次數之賦予的前述各比較資料所進行之比較合格與否判 定結果作計測之計數器手段；和 (2 )在前述初次之資訊檢測時，將合格了的記憶體 位址之上述計數器作1的正數（c〇unt up)，並將此作爲前 述1次突破位址之手段；和 (3)在上述被反覆作了賦予之前述下一次以後的資 訊檢測時’藉由前述位址之置換手段來將前述邏輯積（ -51 - 201137651 AND )演算結果在上述1次突破位址之計數器處進行累積 正數並作爲N( 2以上之比較次數）次突破位址之手段；和 (4)將上述（3)之N次突破位址的位址作輸出之手 段。 進而，係爲一種具備資訊過濾檢測功能的記憶體，其 中’前述進行邏輯積（AND)演算之記憶體的位址之置換 手段，係藉由前述第2比較資料而以位址全範圍來並列地 進行位址之置換。 進而，係爲一種具備資訊過濾檢測功能的記億體，其 中，身爲前述進行邏輯積（AND )演算之記憶體的第2比 較資料之用以與記億體之位址的位址彼此並列地作比較之 資料，係爲下述（1) 、 （2)中之其中一者的資訊檢測用 比較資料：當將前述1次突破位址作爲基準位址，並在前 述下一次以後的資訊檢測時，對於藉由前述第1比較資料 而和記憶體之資料並列地進行合格與否判定並合格了的記 憶體位址、和上述基準位址，此雙方之位址彼此的相對位 置，而用以判定（1 )是否一致的比較資料；（2 )是否存 在於範圍內之比較資料。 進而，此記憶體，係能夠對於用以進行過濾之電路進 行削減或者是將雙重並列邏輯演算作多重化，並且，在邏 輯演算中，不僅是邏輯積或者是邏輯和，而亦能夠進行各 種之邏輯演算的資料和位址之雙重並列邏輯演算。 以上，係結束了對於本發明之記憶體1 2 1、1 2 1 a、 12 lb本身的說明，以下，對於一維、多維空間之資訊檢測 -52- 201137651 的例子作說明。 [實施例1 0 ] 圖1 5 (位址一維配列之資訊檢測例）’例如係將橫軸 作爲時間軸而與位址相對應，且將景氣動向或者是股價、 氣溫等資料在縱軸上作了顯示者，並且係爲根據作爲樣本 而被賦予了的既知之資訊的資料，來對於身爲未知之資訊 的過去之龐大的資料庫中來進行資訊檢測者’此種資訊檢 測’係能夠藉由以將記憶體位址與時間軸附加有關連的方 式而作了配列記憶之資料，而極爲簡單地進行。 作爲以時間軸作爲對象之資訊檢測的另外一例，在聲 音的情況時，係能夠根據壓縮前之取樣時間和與此時間相 對應的聲音資料或者是壓縮聲音資料之每一 AAU (音訊解 碼單位）中的聲音資料，來進行資訊檢測。 作爲能夠進行3値資料之比較的本發明之記億體1 2 1的 其中一例，若是將人的聲音等之頻譜的帶域作類別化並針 對各類別而進行資料化，並設爲1時刻、1位址之量的資料 ，則能夠極爲簡單地作成時間序列之配列。 若是根據此來與樣板（template )音源等進行類似形 態之辨識，則係成爲能夠進行極爲高速之聲音辨識，並能 夠利用在各種之聲音辨識的領域中。 又’此種一維配列資訊的檢測，在將需要莫大之資訊 處理的DN A配列 '基因體之4個鹼基配列之組合解析以高 速來實施的情況時，係極爲有效。 -53- 201137651 同樣的，由文字列之配列所致的解析，亦爲相同。 [實施例1 1] 圖1 6 (位址三維配列之資訊檢測例），係爲對於被配 列在三維空間中之資訊作檢測的情況之例作展示者。 如圖中所示一般，係爲將被配列在三維空間中之特定 的圖案藉由至此爲止所說明之內容相同的方法而檢測出來 的示意圖。 當然的，由於三維空間係可對於人們所存在之實際空 間作表現，因此，係能夠對於可進行三維空間之位置以及 該資料的定量化之全部的三維資訊作適用，並且亦可進而 展開爲加上時間軸的時空間資訊等之多維資訊。 此種三維空間之資訊檢測，係可利用在從原子或分子 程度起直到宇宙空間之一切的配置關係之解析中。 特別是，由於係能夠進行高速的檢測，因此，亦可將 需要進行即時處理之機器人作爲對象，並例如進行將移動 迅速之物體等作爲對象而與多數之樣板畫面作比較一般之 立體圖案辨識、立體物體辨識或者是物體移動追蹤等之中 ，其用途係爲無限。 不論是何種情況，均能夠藉由將記憶體位址與X、Y 、Z之3軸附加對應地來將資料作位址配列並記憶一事，而 容易的實現，並且，除了同一資訊以外，當然的，亦能夠 進行如同在畫像之檢測中所說明之近似資訊或者是類似資 訊的檢測。 -54- 201137651 以上之說明’係爲以一維至多維之空間作爲對象，並 爲於已被作了位址配置之資訊或者是可進行位址配列之資 訊而進行形態辨識者，但是’作爲其中一例，藉由進行像 是將位址分類爲被作了叢集化的資訊群組等並對此進行資 訊檢測等而對於位址和資料的配列作了嚴密設計，亦能夠 與使用有至今爲止的演算法之資訊檢測作倂用。 以上’係爲將一維至多維之資訊檢測出來的情況時之 槪要，但是，若是對於在本發明之記憶體1 2 1、1 2 1 a、 1 2 1 b中之作用的特徵作總括，則係如同下述—般。 使用有本發明之記憶體1 2 1、1 2 1 a、1 2 1 b的資訊之檢 測’係將1位址與其之資料作爲檢測之單位，並藉由輸入 資料之設定方法’來成爲能夠對於從特別嚴密且正確之同 一資訊的檢測起乃至於近似資訊、變形資訊、類似資訊等 之廣範圍的資訊，而高速且確實地檢測出來。 由於用以進行資訊檢測之取樣數或者是取樣點1 1 3之 選擇方法，均係能夠藉由統計之手法來作評價且亦能夠將 取樣自動化，因此，能夠防止取樣數之浪費，並能夠使檢 測時間成爲合理之時間，又，能夠在每一位址（座標）處 而進fr取樣點113之設疋一事，亦爲特徵之一。 另外値得一提者，係在於當在未知之資訊102中並不 存在有成爲對象之資訊的情況時，檢測之中斷亦極爲迅速 〇 此檢測方法，由於係只要滿足一定之條件，則相互之 資訊的大小（容量）上係並沒有限制，因此，係能夠對於 -55- 201137651 所有的資訊作適用，並且亦能夠配合本發明之記憶體1 2 1 、121a、121b之記億體的容量來對於資訊作分割處理。 就算是在將一維資訊至多維資訊以及其他之資訊混合 存在地作了記憶之記億體中，亦能夠進行目的之圖案的辨 識。 又，本方式之另外一個重要特徵，係在於：只要將未 知之資訊1 02適當地作位址配置並記憶在本發明之記憶體 1 2 1、1 2 1 a、1 2 1 b中，則除了取樣點1 1 3以外，係並不需要 進行在其他之圖案辨識中所會進行一般之資訊的加工、特 徵抽出或者是類別化等之資料的前置處理。 進而，本方式，在並不需要進行演算法之開發等的資 訊檢測之實施前的時間一點上，亦係成爲重要的特徵，並 且，檢測方法（比較條件之設定）亦係極爲簡單。 故而，在系統試驗時之試誤法（cut and try )性的調 整亦成爲不必要，便能夠藉由所設定之檢測方法來將所期 待之資訊確實地檢測出來，故而，就算並非特別是資訊檢 測的專家，亦能夠利用本方式而廣泛地利用在各種之資訊 檢測的應用程式中。 在至此爲止的說明中，係以身爲既知之資訊1 〇1 —事 作爲前提並接著對於樣本作採取來檢測出資訊一事爲中心 ，而作了說明，但是，根據人的判斷或者是推測來設定輸 入資料125並進行資訊解析一事，亦爲重要。 在此種情況中，由於身爲最初之輸入資料125的1次比 較資料，係會對於檢測之結果造成重大影響，因此，只要 -56- 201137651 藉由設爲複數之比較資料，或者是在輸入資料125中使其 具備有範圍地來進行檢測，並藉由此1次比較之結果來逐 漸地對於範圍作限定的方法，來進行解析即可。 此種解析’係能夠將天文、氣象、物理、化學、經濟 等之所有領域的資訊預測等之資料解析的時間以及所耗費 之勞力作大幅度的減輕。 本發明之具備資訊過濾功能的記憶體1 2 1、1 2 1 a、 1 2 1 b，係爲將身爲馮諾依曼型電腦之宿命的記憶體之逐次 處理排除，並使記憶體自身具備有智能性之知識而進行資 訊檢測者，而爲將至此爲止之記憶體的常識作了大幅顛覆 者。 故而，不僅是能夠利用在從先前起便被作硏究並被作 利用之畫像辨識、聲音辨識、Ο C R文字辨識、全文檢索、 指紋認證、虹彩認證、機器人之人工智慧的圖案辨識等之 中’亦能夠進行天氣、景氣、股價、分子構造、DNA、基 因體、文字配列等之解析乃至於對新的資訊之發現（資訊 預測）、並且亦能夠廣泛地利用在社會基礎公共設備、產 業用設備、工業用設備、家庭用裝置乃至於至今爲止均未 被作爲對象的未知之領域的資訊檢測中。 [實施例12] 本發明之具備資訊過濾功能的記憶體1 2 1、1 2 1 a、 1 2 1 b ’係亦可作爲需要進行各種之資訊之檢測的人工智慧 之引擎’來利用在知識處理中。 -57- 201137651 例如，係可利用在極爲高度之知識處理中，例如，若 是將本發明之記億體1 2 1、1 2 1 a、1 2 1 b視爲人腦中之大腦 或小腦、並進而區分爲右腦或左腦等一般地來並列且階層 狀地作複數配列，並分別作爲樣板而使其預先記憶物體辨 識之資訊、人物辨識之資訊、文字辨識之資訊、聲音辨識 之資訊、味覺之資訊、觸覺之資訊等等的各種種類之資訊 ，再藉由與即時性所賦予之周圍的畫像或者是聲音等乃至 於各種感測器之資訊作比較，來與人的辨識能力一般地而 對於各種之資訊同時地作辨識，並從該些資訊中來選擇最 適當的行動並實行，利用之方法，係有無限可能。 圖17(使用本發明之記憶體的高度知識處理之例）， 係爲將以上所說明之知識資訊記憶在本發明之記憶體1 2 1 、1 2 1 a、1 2 1 b中並進行知識處理的實施例，並且，經由根 據最適當之行動的結果來對於記憶體121、121a、121b作 更新，亦能夠容易地實現學習效果。 [產業上之利用可能性] 本發明之記億體121、121a、121b，係超越了至今爲 止之先前技術的記憶體之槪念，其用途係極爲廣泛，並爲 建構出一種資訊處理之新的潮流者。 本發明之記憶體1 2 1、1 2 1 a、1 2 1 b，除了聯想記憶體 之外，亦可經由 ASIC ( Application Specific Integrated Circuit)或者是 FPGA ( Field Programmable Gate Array) 來作爲一般性之RAM或者是ROM構成之記憶體來實現之， -58- 201137651 並且’像是將單元基礎（ceU base)之CPU作組入、直接 組入至CCD感測器中、將本發明之記憶體i 2 1、1 2 u、 121b設爲獨自之專用構造的記憶體構成、或者是藉由新形 態之半導體來構成、亦或是賦予其他之功能等，亦爲自由 〇 將來，除了上述一般之由半導體所致的記憶體以外， 亦能夠期待藉由現在正進行硏究之光元件或者是磁性元件 、約瑟夫森（josephson)兀件等之新的元件所致的本發明 之記憶體 121、 121a、 121b。 【圖式簡單說明】 [圖1 ]取樣點之例（實施例1 )。 [圖2 ]由取樣點所致之資訊檢測實施例（實施例2 )。 [圖3]取樣點之評價方法例（實施例3 )。 [圖4]資料以及位址之雙重並列邏輯演算的槪念例。 [圖5 ]具備資訊過濾檢測功能的記億體之例（實施例4 )« [圖6]位址置換手段之第1例（實施例5 )。 [圖7]位址置換之第1示意例。 [圖8]位址置換之第2示意例。 [圖9]位址置換之第3示意例。 [圖10]變形畫像之檢測的槪念（實施例6 )。 [圖1 1]位址置換手段之第2例（實施例7 )。 [圖I2]位址置換手段之第3例（實施例8 )。 -59 - 201137651 [圖1 3 ]資訊過濾檢測電路之削減例（實施例9 )。 [圖1 4]作了多重化之資料以及位址之雙重並列邏輯演 算的槪念例。 [圖15]位址一維配列之資訊檢測例（實施例10)。 [圖16]位址三維配列之資訊檢測例（實施例1 1 )。 [圖17]使用本發明之記億體的高度性知識處理之例（ 實施例1 2 )。 【主要元件符號說明】 1 〇 1 :既知之資訊 102 :未知之資訊 1 0 3 :位址 1 1 1 :像素 1 1 3 :取樣點 1 2 1、1 2 1 a、1 2 1 b :具備資訊過濾檢測功能的記憶體 122 :位址匯流排 123 :資料匯流排 124 :輸出匯流排 125 :輸入資料 1 2 6 :記憶體比較資料 1 2 7 :位址比較資料 128 :重置訊號 129 :比較次數計數器 1 3 1 :位址解碼器 -60- 201137651 1 3 2 :記憶體 1 3 3 :資料比較電路 1 3 4 :位址置換電路 1 3 5 :突破次數計數器 136： Ο R 閘 1 3 7 :條件閘 1 3 8 :突破位址輸出處理電路 1 4 1 :位址置換前合格輸出 142 :位址置換後合格輸出 143 :比較次數訊號 144 :級聯連接 1 5 8 :群組別突破計數器 -61 -201137651 VI. Description of the Invention: [Technical Field] The present invention relates to a memory having an information filtering detection function, a method of using the same, and a device including the memory. [Prior Art] In the era when information is being used for information and can be easily utilized, various problems remain in order to detect and use appropriate information from such huge information materials. In particular, in the technology that is common to information detection such as image recognition, voice recognition, OCR character recognition, full-text search, fingerprint authentication, etc., there is a common technology. Or a similar information (pattern) detected or parsed pattern recognition technology 'this technology, in social basic public equipment, industrial equipment, factory equipment or digital cameras or home appliances, even the latest robots or In various technical fields such as artificial intelligence, they are used as 'indispensable for high-level information processing. However, 'the biggest technical problem in the detection of information as one of the examples is the comparison of the number of combinations (number of searches) when comparing information, usually 'searching for the best algorithm in the object information to compare the number of combinations. (the number of searches) is reduced, or depends on the content of the target information, and a high-speed arithmetic processing device such as a supercomputer is used to find the answer. Further, depending on the situation, there is a need to sacrifice the accuracy of the detection. situation. -5- 201137651 The present invention is to achieve an information detection accuracy that can be used for the long-term problem in the above-mentioned general body pattern recognition or information detection technology, and can compare the number of combinations (number of searches, retrieval time) The maximum amount of memory and the method of using the memory, and the Japanese Patent Application No. 201 0-33376 filed by the same applicant and the same inventor on February 18, 2010. The common management method of the information processing device, the information detection method, the data and the relative relationship between the addresses, and the parallel comparison of the associative memory, the information processing device having the function of common management of information, and the software program thereof All items related to "the relative relationship between the data and the address are combined with the associative memory", and the priority is claimed for this application. In the same manner, Japanese Patent Application No. 2010-472 No. 5, the entire disclosure of the Japanese Patent Application No. 2010- 472, filed on March 4, 2010. In order to make an independent invention from the above-mentioned Japanese Patent Application No. 20 1 0-3 3 3 76, and to change the performance of the invention name to "the semiconductor integrated circuit j having the information filtering detection function, the most characteristic feature of the present invention is The idea of "the logical calculation of the result of the determination of the success or failure of the two-way juxtaposition" is more clarified. Moreover, the scope of the information to be tested is not limited to the two-dimensional image, but also expanded from one-dimensional to multi-dimensional. The information person, and the entire application of this application, claimed to be the priority of this case. In this application, the name of the invention is clearly expressed as "memory with information filtering detection function", and the above two pre-applications are integrated, and the above-mentioned Japanese special wish is 20 1 0-472 15 In addition, the method of reducing the number of filtering circuits for the 201137651 signal, the multiplexing method of the two-parallel logic calculus, and the use of artificial memory as an example of the use of this memory may be added. And changed some of the presentation methods. As described above, pattern recognition or pattern matching is one of the techniques for information detection, and the range thereof is extremely wide. In the invention relating to the shortening of the search time, there is a large number of 'but In the present invention, in order to shorten the detection time, the "single-hand processing of individual memory" which is the fate of the Von Neumann type computer is essentially avoided or is such a memory. The example of the body does not exist. For reference, in the "Data Retrieval Device, Data Compression Device, and Method" of Japanese Laid-Open Patent Publication No. 7-1145:77, it is disclosed that there is a method for repeatedly searching for information by comparing adjacent information with each other. However, the invention of the present application is not only for comparing the adjacent information with each other, but also for the positional relationship between the content of the data and its address as the object of the information of the full memory. Double tied for comparison. [Prior Art Document] [Patent Document 1] [Patent Document 1] JP-A-7-114577 SUMMARY OF INVENTION [Problem to be Solved by the Invention] 201137651 The problem to be solved by the present invention is that One-dimensional or even multi-dimensional information on the location of the address or information that can be assigned in one-dimensional or even multi-dimensional locations, and information for testing (unknown information) and information that becomes the benchmark for testing The address of the mutual information of the (informed information), and the relationship between the information and the address of the address is qualified for the condition, and the information is known from the unknown information. The information is the same information or the information detection of the information-like general pattern recognition or knowledge processing, and realizes a guarantee for the accuracy of the detection, and minimizes the comparison of the data, and makes the memory The body itself is capable of detecting the information, and the non-von Neumann type information is used to detect the memory and establish the method of using the memory. [Means for Solving the Problem] In order to solve the above problem, in the application 1, a memory in which information is stored in each memory address and the information can be read is described, characterized in that The memory system has the following (1) to (4): (1) for comparing the first comparison data that is given from the outside and used for juxtaposition with the data stored in the memory, and for The address of the address of the memory is compared with each other for the second comparison data, and each of the comparison data is used as an input means for input; and (2) by the first comparison data and is memorized in the memory The data is used as a means of juxtancy comparison and judgment on whether or not to be synonymous; and (3) by means of the second comparative data and the means of juxtaposing the addresses of the numerators and judging whether or not the confession is determined; -8 - 201137651 and (4) The logical calculation method of the result of the combination of the above (2) and (3) the result of the combination of the two parties at each address and the logical calculation of the data and the address. . In the second aspect of the present application, the logical calculation of the memory having the information filtering detection function is a logical product (AND) calculation, and the following (1) is provided. (3): (1) At the time of the initial information detection, the above-mentioned comparison data is used to judge the pass or fail along with the data of the memory, and at least one of the qualified memory addresses is used as the address. 1 means of breaking the address and making a memory; and (2) judging whether the pass or fail is judged in parallel with the data of the memory by the new comparative data in the next and subsequent information test, and will pass the test. The memory address is replaced by the address substitution means by the second comparison data; and (3) the logical product of the addresses of (1) and (2) above is broken. (AND) The location of the calculus is the means of output. In the third aspect of the invention, the memory having the information filtering detection function includes the following (1) to (4): (1) individually for each memory address; a counter means for measuring a comparison result of the comparison of the above-mentioned comparison data given by a specific number of times; and (2) at the time of the initial information detection, the above-mentioned memory address is qualified The counter is counted as 1 (count up), and this is used as a means of breaking the address once; and (3) by the above-mentioned information after the above-mentioned information is repeatedly given The replacement means of the address is to perform the above-mentioned logical product (AND) 201137651 calculation result to accumulate a positive number at the counter of the above-mentioned one breakthrough address and as a means of breaking the address of N (more than 2 comparison times); and (4) The address of the N times of the break address of the above (3) is outputted. In the fourth aspect of the invention, the means for replacing the address is that the address is replaced by the entire range of addresses by the second comparison data. In the application item 5, the following characteristics are obtained: that is, the information for comparing the addresses of the address of the genomic body with the second comparison data is juxtaposed to each other as follows ( 1), (2) 2 of the comparison data for information detection: the first break address is used as the reference address' and the information is detected after the next time, for the first comparison The data and the information of the company's data are judged in parallel and the qualified memory address and the reference address, the relative positions of the two addresses are used to determine whether (1) is consistent or not. Information; (2) Whether there is comparative data in the range. In the sixth aspect of the invention, the memory having the information filtering detection function is configured to detect information of at least one of the following (1) to (5). The memory consists of: (1) using sound information as one of the examples as one-dimensional information for being remembered or memorable; (2) using portrait information as one of the examples as two-dimensional information for being remembered or It is information that can be used for memory; (3) using three-dimensional information as one of them as memory for three-dimensional information or for information that can be remembered; (4) using time-space information as one of the examples as a multi-dimensional information to be remembered Or it can be used as memory information; (5) -10- 201137651 The cluster information is used as an example to make the information be remembered or remembered by the group of addresses. In the application item 7, the following characteristics are obtained: that is, the data for juxtancy comparison with the data stored in the memory as the first comparative data is as follows (1)~ (5) At least one of the comparative information for information detection: (1) consistent detection of memory data; (2) size detection of memory data; (3) detection of range of billions of data; (4) memory Individual bit detection comparison; (5) 3値 comparison of memory data. In the application item 8, the first comparative material and the second comparative data are one or both of the following (1) and (2) or both of them are input means. It is used as input: (1) data bus; (2) dedicated input. In the application item 9, the following is a feature that the logical product (AND) calculation described in item 2 of the patent application scope is broken or the N times breakthrough position described in item 3 of the patent application scope is applied. The address of the address is outputted by one of the following (1), (2) or both of the output means: (1) data bus; (3) dedicated output. In the application item 1, the method has the following feature: that is, in the counter means, a means for storing the address of the first break address in the initial information detection is added, and The number of counter means for individual measurement of the memory address is reduced (counter means = number of addresses / η, η is a natural number). -11 - 201137651 In the eleventh application, the processor is equipped with a processor in which the information filtering detection function is provided, and the replacement of the address is performed by the processor. means. In the application item 12, there is a feature that a memory bank segmentation comparison means is provided for performing each of the means (2) to (4) by dividing the range of the memory. . In the application item 13, the means for each of (1) to (4) described in the item 1 is provided in parallel with a plurality of means. In the application item, the memory having the information filtering detection function is incorporated into a semiconductor having another purpose as an example of the CPU and used. In the application item 15, it is characterized in that, in the memory having the information filtering detection function as described in the second item of the patent application, the first information detection and the next time after the next time In the information detection, the foregoing comparison data is given to the memory according to the known information, and the information is stored in the memory by reading the logical product (AND) calculation break address. The information detects the same or similar information as the above-mentioned known information. In the application item 16, the following features are included in the memory having the information filtering detection function described in the third paragraph of the patent application scope, in order to detect the same information or The above-mentioned comparative data of the comparative sample required by the above-mentioned similar information is repeatedly given to the memory, and by reading the count of the aforementioned N times of breakthrough -12-201137651, It is memorized in the information of this memory to detect the same or similar information as the above-mentioned known information. In the application item 17, the following feature is obtained: that is, when the comparison data is extracted, it is required to take out the absolute difference between the data of the adjacent samples and to borrow Information detection is performed by using the sample feature amount obtained by the collection as a specific value or more. In the application item 18, it is characterized in that the information which is the same as or similar to the above-mentioned known information is a pattern (pattren) information. In the application item 19, the following feature is provided: in the step of determining the aforementioned first comparison information at the time of determining the first information of the first breakthrough position, one of the following is performed. Or both parties: to select the most appropriate one from a plurality of samples to be the first comparative sample' and to perform information detection, or to make the first comparative data have a certain range of information to perform information detection. In the application 20, the system has the following feature: that is, the system uses a CPU that can access the memory having the information filtering detection function and read and write data. Information detection. In the application item 21, the feature is that the pattern information is stored as the knowledge information in the memory having the information filtering detection function, and the pattern information is detected. Knowledge processing. In the application item 2 2, there is described a device characterized in that the package contains a memory having the information filtering detection function -13-201137651 as described in the first item of the patent application. [Effects of the Invention] According to the present invention, it is possible to use it as a knowledgeable memory, and it is not only capable of accurately and highly detecting or analyzing all the information and the similarity of the information. It can also be widely used in the technical field of information forecasting or in the highly knowledge processing, and can expect a new information processing process caused by real non-von Neumann information processing. [Embodiment] First, the commemoration of the information detection of the present invention will be described by taking an image of two-dimensional information as an example. In general, when the image of the object (unknown information) and the image of the detection reference (known information) are used to detect the identity of the image, basically, the image (known information) that serves as the detection reference is used. Based on some of the portrait information, the image of the unknown object is searched in a fully consistent manner, and in the case of accuracy, the coordinates of the image are required. As an example of the time spent in the search, 'when a personal computer or a specific pixel of a television picture caused by a digital television signal is used as an object, and a specific material is found from the displayed image, the object is It is about 2 million pixels. This is the first time to expand the bitmap image data of a full screen from the graphic information to the search memory, and the full extent of the temporarily expanded data. When the PU searchs for a simple search such as a specific data at a speed of, for example, an average of 5 η seconds, the search for the first full-scale range (full-screen range) takes 2 million χ 50 η seconds = l 〇 〇m seconds 'Although' is usually after the second time, since the search target is filtered, the search time is shortened. However, it takes about several hundred seconds for the specific image to be specified. Therefore, when When it is necessary to search a large number of images on one screen, the search time cannot be ignored regardless of how high-speed processing is performed. In addition, the above description is only the case of the same image. When it is assumed that the size of the image is changed or rotated, since the calculation of the coordinate transformation is required to be repeated, the processing time becomes the number of cases. A hundred times or even thousands of times or even longer, such a search is difficult to achieve. The above situation can be said to be an inevitable fate of information detection by a Von Neumann computer that needs to perform a CPU search for each address of the memory. Due to the technical background of the general search time described above, the mainstream technology of the current search is a search for the similarity between the portraits due to the clustering of the feature data extracted from the features of the portrait, and It is used in a wide range of fields such as face recognition of digital cameras or smile recognition, and even voice recognition. However, the search ability such as the accuracy of the detection or the time of the search, the information that can be detected, etc., depends on the hand-extracted hand--15-201137651 method or the clustering method. change. In addition, in the field of use of image search, there is a case where there is a case where a false positive rate causes a fatal error. Therefore, there is still a need for a portrait search that requires similarity and similarity. As described above, the accuracy of the pursuit of the image search and the shortening of the time are incompatible with each other in terms of purpose. However, first, the description of the same image is surely performed in the image search. As described above, the present invention is generally required for authenticity. In principle, the information is detected by using each coordinate (address) as a comparison detection target. Hereinafter, in order to achieve this, it is impossible or Explain the type of information that is missing and its resolution. There are various types of information for portraits. However, the two types of portrait information are roughly described as an example. First, when the data from the frame buffer (graphic memory) of the image to be displayed is used as the image information, in general, in the case of color, the total of R, G, and B is 16 bits to 64 bits. Hold the information for the length of the data. The R, G, and B color signals can be used as a target. However, as an example of the effective image detection, the computer or the image device is buffered at about 2 million pixels. By taking four bits of R, G, and B (three of the 16 combinations) and using one pixel data, it is possible to detect the image with good precision even in the case of any color. In this case, the combination of the colors is 12bit, 4096 combinations, -16- 201137651 When the color on the screen is uneven, the probability that one color exists on the screen is 2 million / 4096. 4 48 8 pixels (address). Second, as an example of this, one block of a plurality of compressed image data such as JPEG (Joint Photographic Experts Group) or MPEG (Moving Picture Experts Group) can be used (for example, 8 X 8 pixels) As a coordinate, the DC (DC) component data of the DTC (Discrete Cosine Transform) of the brightness of the block or the color difference signal is directly used as the data of the coordinate, or for other information (such as vector) Information) for use. In the case of this DCT, since it is a block unit, the number (address) of the coordinates can be reduced (as an example, for example, 1 / 6 4 ) as compared with the pixel. Of course, in any case, it is better to use high resolution. However, since the memory capacity is also large, it is only necessary to select the necessary number of bits from the LSB side of the above-mentioned general quantized data. And as a portrait information material. Hereinafter, the above-described pixels are targeted, and the image data obtained from the combined data of R, G, and B of 2,200,000 (bits) and 1 2b, 4,096 are used as portrait information. To illustrate the case of detecting this. [Embodiment 1] Fig. 1 (example of sampling point) is for 丨, 2, 3, 4 from the upper left corner of the screen to the right corner 隅. . . . . η η pixels 111 of the capital -17- 201137651 based on the address of the memory 1 0 3 1, 2, 3, 4. . . . . When the order of η is used to match the memory, The sampling point of the image (the known information 1 〇 1) that is the basis of the test is 1 1 3 as the exhibitor ❶ the test reference image A, It is based on a smaller image. And in the field of the image (known information 丨〇丨) that is the basis of the test, the center is centered on the coordinate y〇 ' x〇. In the case where the total of 25 coordinates are taken as the sampling point 1 1 3 at equal intervals, up, down, left, and right, the automatic arrangement is performed. In this case, it is 33 pixels 111 each at the XY axis. That is, a total of 1〇89 pixels 1 11 is taken as an object. Even if the size of the image of the detection standard is increased, There is also no problem at all. The numbers from 1 to 2 5 shown in the figure, Department represents the order in which the samples are compared, In this case, the center is used as the base point. And set to detect from the diagonal sampling order from the sampling point 1 1 3 farther from the center toward the closer sampling point u 3 'but' the arrangement or the comparison order, The system is not limited to this, For details of this matter, It will be described later. Detecting the reference image B, Are not arranged at equal intervals as described above. However, an example in which the sampling points 1 1 3 of i 22 are manually set in accordance with the image of the detection reference is used. So by manually setting the sampling point 1 1 3 and detecting it, In order to make use of the high feature recognition ability of human beings, In the case of specifying a characteristic point or a characteristic range and seeking differentiation from other pixels, Is valid. By using one -18-201137651 coordinate as the detection method of the detection unit in order to achieve the accuracy of the image detection, it is possible to set the sampling point at each coordinate point. Other searches that use multiple areas as a feature point in a certain area, In the detection mode, The system is not able to achieve one of the features of the present-style approach. [Embodiment 2] Fig. 2 (Information detection example by sampling point), A description will be given of a case where an image of the detection reference image A of Fig. 1 described above is used as a reference for detection and an image is detected. The department shows: An image that becomes the basis of the test (known information 1 〇 1 ), And the portrait of the object to be detected (unknown information 1 〇 2). The group A shown in the map of the unknown information 102 is a case where it is uniform in one comparison but becomes N G in two comparisons. Group B is a case where N G is compared in 11 comparisons. Group C is a case where NG is compared at 22 times. Group D is the case where all sampling points 1 1 3 are consistent. Such detection, It is only necessary to read and compare each other at each address as long as the data of the coordinates described previously. In this picture, Although it is explained by the image area from the group A to the group E which are completely separated in position, however, In the case of portraits, usually, As the coordinates of the same material, Whether it is adjacent or centralized, and if the resolution is lower, the tendency will be more significant. but, As long as you make a correct comparison of the relative positional relationship between each other, There will be no problems. In this case, Although in chance, The area of the group D is also likely to be judged to be the same image as the screen that becomes the reference. but, By repeating the sampling point 113 in addition to -19-201137651, the consistency of all the pixels of the mutual portraits is confirmed again. To be able to exclude similar images and ensure that they are identical portraits. This method, When detecting the size of the object to be imaged, The system is most suitable. By using the general verification method described above, When the sampling point Π 3 is partially concentrated and becomes inconsistent as in the group E, It is also possible to judge the portion of the consistent portrait as a deformed portrait to which a change is applied in a part of the portrait. also, In the case of group C, Can also be judged as a similar portrait, In response to these situations, It will be described later. As the above, the first coordinate of the sampling point 113 at a certain place is used as a reference. And determine the relative position between the coordinates of the subsequent coordinates, Then, there is a method in which the contradictory coordinates do not exist as the candidate coordinates in the relative position, It is a common means of information retrieval in pattern recognition, etc. A method and apparatus for efficiently performing such a combined search and detecting an image (information) at a high speed, Is the essence of the invention, Regarding its details, It will be described later. In the method of detecting the portrait described above, Of particular importance, In the selection method of the sampling point 113, One of them is in the scope of certain portrait data. The other one is the degree of change in the image data. For example, in the case of a black portrait portion or a white portrait portion where no change exists, Or to specify a portrait with less change, Or when, for example, when there are only images having a feature such as a portrait with text information, it is specified. The detection of this portrait becomes difficult. The following 'shows the effectiveness of the detection of images and portraits, The existence of trust -20- 201137651 The sampling point with deep relationship 1 1 3 considerations. If one of the coordinates on the portrait is used as a reference, Then the probability that the coordinates of the coordinates adjacent to the coordinates are the same as or similar to the reference coordinates, that is, the correlation, becomes higher. And as the coordinates move away, the correlation becomes lower. Therefore, In the same way as the sample of Figure 1, each time it is confirmed from the farther sampling point 1 1 3, Consistent, The probability of inconsistent judgment becoming faster and becoming an efficient test is higher. Therefore, the probability of existence of the same coordinate group caused by the total number of sampling points 1 1 3 that have been completely decentralized, It is an index that accumulates the number of sampling points 113 on the bit number of the data of the sampling points 113. For example, 'the first coordinate is R, G, When B is composed of 4 bits, the number of combinations is 2 to 2 (2 K). but, If the sampling point 113 is 10 places, Then the number of combinations of 2 to the power of 120 is in actual operation. Become a probability combination equivalent to infinity, As long as each other is not characterized by a monotonous portrait, If all of the sampling points 1 1 3 are the same, It can be judged as the same - portrait. However, "because there are cases where a limited image of a limited range is targeted" or a black and white image such as a text is also targeted. Therefore, it is not possible to take the above-mentioned general sampling point 113 which has been completely decentralized. Therefore, it is necessary to perform the evaluation of the identification ability of the sampling point 1丨3 described below and take, for example, an alarm or enlarge the image area. It is also possible to take appropriate measures such as adding the sampling point 1 1 3. -21 - 201137651 [Embodiment 3] Fig. 3 (Example of evaluation method of sampling points), An example of evaluating the recognition ability of the sampling point 1 1 3 in the case of the detection reference image A of FIG. And for the sampling point 113 in which the coordinate 1 is used as the detection reference coordinate and the total is 25 as the coordinate 25, To set the adjacent four sampling points 113 as one group. On the other hand, there are a total of 16 groups of adjacent sampling points of A to P. As an example, In group A, Contains 2 1〇, 14, 4 of the 18 sampling points 113, In group B, Contains 10 6, 18. 4 of 22 sampling points U 3, The same is true below. at this time, 4 sampling points of each group A to P group 13. In each group, Whether in brightness information or color information, There are differences in the coordinates of the data, Due to the size of the system and features, That is, it is related to the size of the sample feature quantity. therefore, By taking 6 combinations of 2 combinations from the 4 samples, That is 2 -1 〇, 2 - 1 4 , 2-18, 10-14, 10-18, 14-18, The absolute amount of the difference between the data to be taken out, And the total of the six combinations and the collection of the whole (16 groups) are obtained. It can be set as a specific defect of the feature quantity of the group. When R, as in this case G, When B made a composite data, As long as they are evaluated independently. also, When grasping the ability to recognize the same number of samples, the feature quantity of each of the group A to the group P can be aggregated and divided by the group of six. The average feature quantity obtained is used as a measure of the size of the feature quantity (special 値). of course, If the feature quantity of this sample is small, Because it will affect the identification ability of sampling point 1 1 3 for -22-201137651, therefore, When specifying the reference image to be detected and determining its portrait area, As long as it is made more than the benchmark, generally It is only necessary to increase the number of sampling points 1 1 3 or to increase the area of the image, etc., as necessary. Needless to say, There may be the opposite situation, Evaluation method for this sampling point 1 1 3, In the case of the number of search processing (time) of the detection of the image, the number of times of retrieval (time) is reasonable, It is important. the above content, When the sampling point 113 is determined manually, Of course it is also effective. In this case, Although it is an example of evaluating the ability to recognize two-dimensional information as an object, but, The idea of being able to extract the difference between the one-dimensional to multi-dimensional adjacent samples and the data of them, To determine the feature amount. and then, Depending on the type of information that is targeted, If the basis of the sample is defined and judged solely based on the characteristics of the information, It is a test that enables more accurate information. So far, It is directed to the detection of information that is indispensable for the authenticity of the invention. The image of the two-dimensional information is described as an object. But the basic content of this idea, For information on other dimensions, Also common. herein, An explanation is given of the arrangement of information for realizing the memory of the present invention in the memory. One-dimensional information, It is information that is continuously remembered on the memory address. Two-dimensional information, It is like the bit -23-201137651 of the 1st to η of Fig. 1 and the maximum coordinate number of the individual dimension is used as the basic condition of the column and is transformed into a table (tab 1 e ) and used as a one-dimensional memory bit. Addresses are continuously remembered or can be used as information for the billions. It is composed of a memory that allows the size of the memory and the memory capacity to correspond to their respective purposes. The same is true when 3D or even multidimensional information is memorized on the memory address. Information about this, The maximum number of coordinates of the individual dimensions is used as the basic condition of the matching and is transformed into the same information as above. And the memory composition is also the same. Therefore, This information, If given the coordinate data corresponding to each dimension, It is possible to specify the address corresponding to the coordinate data given by the maximum number of coordinates of each dimension according to the basic condition of the body, The relative position of the coordinates or the range of coordinates can also be specified. of course, In the case of high-dimensional information, The same, When it is possible to directly know the address of the object, The system can be specified directly with the address or relative address. the above, For the purpose of confirming and making up the record again, This address is assigned to the method. Is a general information matching method that is usually performed. This generality is a method for compiling information about the billions of bodies. Since it is a basic information arrangement for implementing the present invention, therefore, It is extremely easy to use. after that, Based on the idea so far, The case where the memory 1 1 1 having the information filtering detection function of the present invention is applied to the associative memory will be described. -24- 201137651 Associative memory, It is used in a memory that is particularly important in the information processing device, such as cache memory or communication data processing, and requires high-speed processing. also, It has been extensively studied as a search device for the purpose of detecting short-distance search (similarity distance search) for similar images by clustering techniques, for example. then, For the description of the associative memory. Associative memory used in high-speed data processing, In addition to the usual memory functions, It is also possible to externally and simultaneously compare (parallel) the data to the full memory. It is a device that is very convenient for searching information by reading the address of the qualified memory. If you give an example of this parallel processing, For example, prepare a seat (address) in a venue where a large number of people gather and ask people to sit down. And when it is set as a card (material) which allows these people to freely select a color of a favorite color, for example, when investigating a red card (material), In the case of normal memory, It is necessary to investigate all the people by successive comparisons according to the order of the seats, etc. In contrast, In the case of '丨η' of the associative memory, for example, the person holding the red card should raise the hand (parallel comparison) and confirm the seat (address) (the output processing). therefore, It is capable of making extremely high speed determinations. So this kind of ‘Len’s mind, When juxtaposing a large amount of information and looking for only the necessary information from it, It becomes a convenient device. -25- 201137651 Although associative memory has various advantages, but, As one of the weaknesses in the composition, When a memory device that reads and writes data due to a data bus and an address bus is performed, Even if it is possible to compare data from the outside (in this case, It is red) and its qualification is judged at the same time. When the qualified address is plural, The address cannot be output at one time. In order to solve this, As long as it has the priority function at the output, And the qualified memory is sequentially outputted to the address of the qualified memory. but, If it is not the number of addresses that have been fully filtered, It will also take time to read. In the usual case, This address is read out. And the following comparison conditions are given to the address that has been read, after that, Based on this condition, filtering is performed by successive processing. In the previous example, 2 million pixels (address) is 12 bits. In the case of a combination of 4 and 96 types of data, It is necessary to have an average of 48 8 addresses as objects. And repeat the subsequent conditions, usually, Most of them are pixels outside the object (which will not remain). And for unnecessary processing, After 3 times, It is also the same. Therefore, The device of the present invention having the information filtering detection function is completely excluded from such sequential processing. And then have: Among those who hold the red card, the person who makes the relationship (the relationship of the address) is the same as the person who is in the same position (for example, the neighbors are each other or the person holding the red card before and after) The function of putting your hand down is to give the relationship between the information and its address, That is to say, all the -26-201137651 memory (the person in the seat) is given the comparison condition (the color of the card, Relative relationship of seats), To judge the eligibility of the material that is memorized (the color of the card) and the positional relationship between the addresses (the relative relationship of the seats), Then, the address that has passed the logical product (AND) calculation due to the result of the juxtaposition of the two parties is filtered and detected (the neighbors are each other or the person holding the red card seat before and after) And the function of outputting it, That is, This memory, Is a memory capable of double juxtaposition pass or fail. Furthermore, a memory capable of performing various double parallel logic calculations is realized. Figure 4 (The commemoration of the double parallel logic calculus of the data and the address), It is the complication of the logical product (AN D ) calculation of the result of the double juxtaposition of the data of the memory described above and its address. That is, Is used to compare the content of the information of each address, And comparing each address, And juxtaposed (double juxtaposed) for the pass or fail judgment, And based on the result of the pass or fail judgment, Logical product (AND) calculus for filtering information, Furthermore, proceeding side by side 〇 the result of this calculation, No matter what form it is used for, it can be used. 0 As above, Even if it is filtered once, Can also get a great filtering effect, but, If this is done in parallel, the double parallel logic calculation is carried out continuously and repeatedly. This makes the ideal information filtering detection possible. In order to realize the above idea, the address comparison failure circuit of the address comparison in the lower part of FIG. 4 is expressed as a mourning, Although it is easy, -27- 201137651 However, In the usual idea, Because of how to compare the addresses, And what is the comparison with the address of the person, Department is not necessarily, therefore, To actually circuit this mourning into logic, It is not easy. E.g, Although it can be considered: The address to be compared and remaining as described so far, As a breakthrough address, And use this as a benchmark, To compare with each address, but, In this method, The same is true as explained above. If it is assumed that there are 48 8 addresses in the 1st breakthrough address system, Therefore, since it is necessary to form a combined parallel address comparison circuit caused by the combination of all the addresses and the 48 8 addresses, therefore, It is a very large scale. If it is a small number of memory addresses, However, the address comparison can be realized by the above configuration. but, In this embodiment, It is also possible to realize the logical composition of Fig. 4 by a simple circuit configuration as much as possible in the case of a large-scale memory. Pay attention to the following situations: That is, By defining one break address as the address of the reference origin used for comparison each time, Each subsequent comparison address (coordinate), The first break address and the relative address (coordinate) of each of these will be the same relative position (address) in each comparison. That is, The optimum solution for this logic circuit is obtained by comparing the relative relationship between the address of each of the breaks and the subsequent mutual address. in particular, The system can be configured as follows: That is, For each of the sampling points 113 shown in Figure 1 previously described, Use 1 breakthrough address as the reference origin coordinate. And set a certain relative offset or range in the comparison address 丨〇 3, And for each! Sub-burst -28- 201137651 Is the information of the address of the target of the broken address 1 〇 3 qualified for this? If it is qualified, Then set this as the break address. and then, By setting a counter for recording the number of breakthroughs at each address 103, And it is set to count up the number of breakthroughs at the first breakthrough address, which is the base origin. It is possible to determine the address of the memory of the highest number of breakthroughs (N times) in one breakthrough address as N times of breakthrough addresses. To make continuous filtering logic possible, Even the composition of a very simple logic circuit, It can also be formed to satisfy the composition of all the original purposes. It is possible to implement a device that overcomes one of the problems of information processing for many years. Although it will be described later, but, Figure 4 is a double parallel logic calculus, The system is not limited to only the logical product (AND) calculus. also, The way in which the results of this calculation are used, Also free. [Embodiment 4] Fig. 5 (having a memory example with information filtering detection function), It is the memory of the present invention which will be associated with the above contents! The function of 2〗 is to be displayed on the basis of associative memory. The details of the timing of data processing, etc. are omitted. It is only for the memorandum of the matter related to the present invention. Here is a memorable body with information filtering detection function! 2 1 (the following is also recorded as the memory of the present invention), Is connected to the address bus 1 22 Data bus I 23, It is also a structure that enables the exchange of information with the outside. -29- 201137651 Therefore, the memory of 1~n is a billion body 1 32, The address 丨~η is selected by the address decoder 131 of the address bus 122, And can write data from the data bus 1 2 3, Reading the input data 125' is for the memory 121 of the present invention to be provided with information for detecting information, The memory comparison data 1 of the first data is the data used for comparing the data of the memory from the outside. 'The data between the input data 125 and the memory 132 of the memory 1 to η is Qualified or not, The data comparison circuit 1 3 3 is used for the determination, When qualified, This result is output as the qualified output 1 4 1 before the address replacement. For the address comparison data 127 which is the second input data 125, And a address replacement circuit 134, It will be described later. Breakthrough counter 135, The counter is calculated by replacing the qualified output 142 by the address and using the number of times of the data comparison circuit 1 3 3 as the number of breakthroughs. This breakthrough number counter 135, The system has a consistent output function between the comparison count counters 129 and the comparison number of times 143 for counting the number of comparisons between the information. Its output, Is connected to the OR gate 136 and the inhibitor gate 137. And by cascading the connection of the signal 44 from a smaller address, On the other hand, the priority processing for the address which is the smallest of the number of breakout counters 135 having the number of breakthrough times is 135, and the priority processing of outputting only one address is performed. In addition, 1 2 8 is the reset signal. The break address output processing circuit 138' performs processing for transmitting the address of the priority output to the output bus 1 24, And the completion counter 135- 201137651 address of the breakthrough number counter 135 is cleared, And become: After that, if there are other N breakthrough breakthrough counters 135, You can use its address as the next priority output. Then, the address of the N times of breakthrough is sequentially sent to the outside through the output bus 1 2 4 . The dedicated output form of the dedicated bus output of this example, For the only example, the output can be sent directly to the data bus 1 2 3 . Therefore, if it is based on this composition, Then, the address (coordinate) of the burst number counter 135 with the highest number of breakthroughs (N times) becomes Winner (N times breakout address)', and the address is outputted in the order of small and large addresses. [Embodiment 5] Fig. 6 (first example of the address replacement circuit), It is an extremely simple logic circuit configuration' and is a basic tribute to the address replacement circuit 134 that is the means to implement the present invention. Address replacement circuit 134, It is set between the data comparison circuit 133 and the burst number counter 1 3 5 , This address replacement circuit 1 3 4, For the sake of comparison in each sample, Breaking the address at the first time of the purpose, The output of the breakthrough is used as the result of the double parallel logical product calculation as the cumulative addition setting. In the case of this example, The system is composed of: The address comparison data 127 of the second data, which is the input data 125 described earlier, can be used to replace the address before the address is replaced by the relative address comparison data. Figure 6, i, j, k is transformed into χγ axis coordinate data, And shifting the transformed qualified output to the amount of the relative address, Then, as the address replacement, the qualified output 142 is input at the breakthrough address counter 1 3 5 (1 break address) of the address. -31 - 201137651 The qualified output is input as the breakthrough output. That is, Qualified output after address substitution 1 42, When the relative address condition of the address of the one-breaking address is qualified, It is input as a breakthrough output to the 1st breakthrough address. of course, It may not be a coordinate information. Instead, the relative address comparison data is directly specified by the relative address and shifted by the amount of the relative address. The first described above Input of the second data, To give an assignment from the data bus 123 or to give it from a dedicated input, It is free to store the pixel data of the previously described portrait or the equivalent information in the memory 1 1 1 of the present invention constituted as described above, An example of the case where the image is detected based on the unknown information 102 is explained. In the memory 132 of the memory 1 to η of FIG. 5 previously described, The pixel data of the image is written to the address corresponding to the respective coordinates. Comparison count counter 129 and all break count counters 135, All the departments were cleared and became 〇, after that, The comparison count counter 1 29 is counted up in each comparison. First of all, As a comparison, Pixel data of sample 1 is given as input data 125 to the comparison data 126. Parallel determination of all memory, The address comparison circuit 133 replaces the qualified output 141 as a one-time qualified output for output. This 1 qualifying output, And is not replaced by an address, And as the address replacement qualified output 142 is directly added to the input of the breakthrough number counter 135, The 计数器 of the counter of the address -32-201137651 that has been broken is set to 1 ’, which is the 1st breakthrough address. As above, In one comparison, The second information is not required. As explained earlier, The number of occurrences of the average breakout address is 488 (Fig. 6, i, j, k) ’ this number appears, Since it is a hypothetical number, So ‘even if it’s more or less, it’s fine, the following , It is also the same. In the case of Figure 2, at this time, After the breakthrough counter 1 3 5 becomes the memory address of 1, Is the group A Group B Group 1 The coordinates of the group 1 and the group 5 are the winners of Winner (1 breakout address). This is an address (coordinate) that is the focus of the description that will be described later. Then 'as a comparison 2', by designating the pixel data of the sample 2 as the comparison data 126, The other locations were again averaged as 2 qualifying outputs and 48 8 were selected. and then, By specifying the difference between the addresses of sample 1 and sample 2 as the address, comparing the relative address comparison data of the data 1 2 7 , In the newly selected 488 addresses, the relative relationship with the previously described one break address is qualified. That is, the group relationship is established. The breakout counter 1 3 5 (1) is used to shift the address (coordinate) corresponding to the difference by the position permutation circuit 1 34 shown in FIG. 6 and to perform the relative position of the shift transform. Breakthrough address) After the address is replaced, the qualified output 1 42 is used as the breakthrough output for the addition input. That is, The breakthrough number counter 135, which is originally the coordinate (address) of the group A to E of FIG. 2, is counted as a count up. but, -33- 201137651 At the breakout counter 135 (1 breakout address) of the candidate coordinates surviving in the judgment of sample 1, The offset of the relative address is applied in such a manner as to continue to count up. And as a breakthrough output of 2 times, it gives a breakthrough input for 1 breakthrough address. In the case of Figure 2 previously described, at this time, Breakthrough counter 1 3 5 becomes the 1 breakout address of 2, Is the group B Group C 1, The four places of group D and the group 1 of group E are maintained as coordinates. Group 1 of Group A is not counted as (count up), It was removed from the alternate. The above content, And the position of the address of the object (which is equivalent to the address of the second sample) is determined by using the first break address as a reference to determine whether it exists at the destination (relative address comparison data). And the logical product (AND) calculation is performed in parallel with the qualified address caused by the comparison of the data of 2 times, and the result is input as the breakthrough output to the breakthrough number counter 135 of the 1 breakthrough address. . The data of the memory 1 32 is similarly compared in the same order and the relative address between the other samples is taken as the reference 1 sample. Read in as a pair of input data 125, And in each address group, the memory that breaks through is concentrated in one breakthrough address and is positive. By this, It is possible to continuously perform the filtering of the breakthrough address (Winner candidate) once. Therefore, In Figure 2, Has broken through to the last sample 25 (N is 25) of the breakthrough address, There is only the coordinate of group D 1 (25 Winners). The number of breakout counters for this coordinate (address) is 135, Become 25, This breakthrough number counter 135, The ratio of the comparison counter 129 is -34- 201137651. The comparison signal 143 is consistent. The output, It is input to the subsequent 〇R gate 136 and condition gate 137. Figure 7 (the first schematic example of address replacement), Qualified for both the "qualified information" and the "qualified relationship of the address" as described so far. That is, the content of the breakthrough caused by the double parallel logical product calculus. As a schematic diagram, as shown in Figure 7, In the coordinates of the picture, A total of 6 breakout addresses from A to F are displayed by the first comparison. This address is replaced, Relatively speaking for all addresses as objects, but, 1 break address A~F, As in each location group that becomes the object of comparison, Looking at the relative coordinates of the two comparisons that were compared for comparison by the telescope, And when it is the qualified output of 2 qualified addresses (in this picture, When it is marked with a black circle, Then take this as a breakthrough output, Just as the same as the replacement action. 3 comparisons, The same, By looking at the relative coordinates of the comparison by the telescope, And the qualified output of 3 qualified addresses (in this figure, It is marked by a black triangle. It is captured as a breakthrough output. after that , The same replacement is also performed. In the case of this example, The counter of E is updated to 2, and then, The counter of B is updated to 3. Figure 8 (the second schematic example of address substitution), Is the A shown in Figure 7 above, The indication of the coordinates of the two first-breaking addresses of B is explained as an address. As shown in Figure 8, in general, 1 break address A, B, The schematic is like -35-201137651 by the telescope to look at the data comparison circuit from sample 2 up to sample 25, And when there is a qualifier in the data comparison circuit that has been gazing, Take it as a breakthrough output. Of course, Switching of the telescope, By comparing the relative address of the data used as a comparison of the relative positional relationship of the addresses, It is set in every time. Actually, With 1 break address A, There is no relative relationship between the addresses of B, 2 times, 3 times, Qualified output due to N comparisons, It is also made a positive number at the address that has been displaced relatively. but, If the sample is not suitable and meaningful, The system will be scattered every time, A qualified output system will not be concentrated at a specific address. This is because the special relationship (pattern) that is "the same as the specific part of the image (information) of the sample) is not established. In the indication, 1 break address, Constantly maintains the superiority of the count (initially 1), And as a dominator's general position to have the right to have a group of samples associated with the relative address of the 1st breakthrough address and to pass the qualified output as a breakthrough output. Figure 9 (the third schematic example of address substitution), It is used as an example for the replacement of the address in the actual 2-dimensional array address. Table A, B, It is used as a display for the address (coordinate) 1 to 100 before the address replacement. twenty four, 50. 67. The four addresses (coordinates) of 72 are the first breakthrough address. Table A, In the case of comparing the address of the second comparison address with the address of the relative address - 22 -36-201137651 data comparison address case] The coordinates of the target of 7 2 ', It is outside the object. Table B, It is the case where the comparison address is used as the data comparison address with the address of the relative address + 3 1 'the address of the destination address 50 and 72'. It is outside the object. Table C, Is to make the table A - 2 2 address shifter ' 2 4, 50. 6 7 times to break through the address, The ability to properly obtain the results of the data of the respective targets, And if it is a qualified result, Then, it is possible to make a positive number (permutation count) as the burst output. Table D, Is to make Table B a + 31 address shifter, twenty four, 67 times the breakthrough address, Being able to normally obtain the results of the data of the respective targets, or not, And if it is a qualified result, It is possible to make a positive number (displacement count) as a breakthrough output. Repeat the above operations a specific number of times, The coordinate position of the target of the comparison object is the normal I-break address. The department can survive until the end. In the above-mentioned contents explained by various examples, And by "address replacement (address replacement), To use the breakout address as the benchmark 'after' for whether there is the same data (qualified data) as the sample point 1 1 3 that is compared at each time and then exists at the destination (relative address comparison) Both) And pass the pass or fail judgment continuously and in parallel. And then perform the logical product (AND) calculus in parallel, And the result is equal to the number of breakout counters 1 3 5 that are output to the breakout address once every time. -37- 201137651 Address Replacement Circuit 1 34, In order to make the juxtaposition of the lower part of Figure 4 pass or not, the judgment is integrated with the parallel logical calculus. It is a logic circuit that has excellent efficiency in two functions (that is, an equivalent double parallel logic (AND) calculation method). Final Results, As long as this is the N times of breaking the address as the number of comparisons, Readout by the breakthrough address output processing circuit 1 38 and the output bus 1 24 It is possible to specify the address of the group containing the information of the N breakthrough addresses. That is to say, pattern recognition. By comparing the comparison number counter 1 29 as a presettable counter, the comparison number signal 1 43 is specified, The address of the counter (N times of breakthrough address) of any count 値 can also be read as an example of the address replacement circuit 134 that is read out on the way. As shown in Figure 6, Prepare a scratchpad for address translation, And the address obtained by the data comparison circuit 133 is replaced by the pre-qualified output 1 4 1, Relatively making a mobile permutation by comparing the coordinates of the data with the address, And set to address replacement and the output is 14.2 This register operation, Since the system is a relative shift of all addresses, therefore, Even through the data transfer method of adding and subtracting calculus, Or via the simplest shift register with the data length of the address, It can also be easily implemented. The above address replacement method using the scratchpad, Is one of the examples used to explain, It can also be implemented by other methods such as replacement (address replacement) using an address decoder directly, etc. Similarly, Address replacement circuit 134 and breakthrough number counter 135, -38- 201137651 Although it is an indispensable means in realizing the information filtering detection of the present invention, but, Not limited to this configuration, Other methods, Come to implement individually at each address. a method for determining the pass or fail of the batch in relation to the relative relationship between the data and its address, It is judged that the pass condition of the data comparison condition of all the memories 1 to 2 of 1 to η shown in FIG. 4 and the qualification of all the address comparison conditions are judged in parallel, and further, The logical product calculus is carried out side by side. The matter is equivalent. therefore, In principle, It is necessary to process the addresses of the von Neumann-type information processing that treats individual addresses as objects one by one. Therefore, The unknown image of 2 million pixels described so far is set to a resolution of 12 bits. When the number of data groups is 4096 groups, The comparison of the number of data required for the detection of the same image from an unknown portrait is usually converged at 2 or 3 times. Even if it is the largest, it is possible to find the image of the purpose by the number of comparisons of the number of samples (25 in this case). Further, this way, Since it is as shown in Figure 7, For the full-coordinate pattern matching by the input data with the one-stop address A to F as the reference origin, the comparison is performed every time (all the objects are compared as objects), Therefore, it is possible to implement an image that is similar to the case of a part of the image (hereinafter, Detection called an approximate image) 〇 For example, In the case of Figure 2, after the end of the comparison of 25 times, The counter of the coordinate of group 1 is 1 and the system is 1. The counter of the group B 1 is -1 - the ratio of the 201137651 is 1 〇. The coordinate of the counter of the group c 1 is 2 1, The coordinate of the counter of the group D 1 is 25, The coordinate of the counter of the group E 1 is 22. That is, The counter is the address of the high coordinate. The possibility that it is a part of a portrait that lacks a portrait or an approximate portrait is high. In this case, Even if it is a breakthrough address, it does not matter. As long as the counter 値 is more than a certain number (for example, 20 times), the coordinates are read in advance. And specific (in this case, After the comparison of 25 times), A detailed judgment can be made for the coordinates of the periphery as needed. Therefore, This way, It is effective not only when the same image is detected at a high speed, but also when it is detected at a high speed. When detecting an approximate portrait based on a certain definition, Also effective. also, By further developing the idea of this address replacement (address replacement), It is also possible to enlarge or reduce the portrait (hereinafter, The image called the deformed image is detected by the minimum number of detections. [Embodiment 6] Fig. 1 〇 (The mourning of the detection of an deformed image), For the comparison, the image is enlarged or rotated, It is a valid method for detecting when a deformed image with a change is applied as a premise on the data. Figure 10, It is the one breakthrough address of the unknown portrait of the one who passed the comparison of one sample. The sampling point 113 of the known portrait is superimposed. -40- 201137651 In this example, For the existence of the original dimension of the coordinate 1 to make the size of the unknown image at X, 2 times on the γ axis, The situation is shown in the case of the possibility of expansion of the system. When the deformed image found is in this case, the coordinates of the 2~2 5 of the sample corresponding to the shape image, Should be inside the circle shown in the figure, Therefore, As long as the circle is included as the image detection range of this coordinate 1. Therefore, The position of the address (coordinate) up to this point is expanded, By judging whether or not there is a seat having the same information as the sample in the circumference specified by the reference origin of the coordinate 1, There is no number, It is only a simple judgment that the number of samples is determined and specified (in this example, Is the coordinate of the sample equivalent in this range, It becomes an image that can be detected. In this case, The same, By determining the number of special samples of the sample and the identification ability to be a certain benchmark, the object is within a certain range, Its authenticity is improved. In this case, the same is true. As long as the output of each pass (breakthrough) of each address is concentrated at the first time of the destination and used as a means of memory addition (in this example, Is a method for making the decision to accumulate at the breakout counter 135 of the coordinate of 1), For such deformed portraits, It is also possible to detect an image by setting the number of samples to the most data comparison times. Point as medium (as a painting, And the coordinate model of the commemoration of the coordinate of the existence of the association (none), Also 25) is enough to deform the levy and set the range of data and the result of breaking the address is even larger and the least -41 - 201137651 [Example 7] Figure 22 (The second example of the address replacement means ), In order to achieve the above ideas, The address conversion described in FIG. 6 is shifted from the 1-to-1 address and developed to be the range of the address. And input this from the outside as the address range comparison data to the address comparison data 127, With this, To qualify for this address, j, Before the address of k is replaced, the qualified output is 1 4 1, Read as the address range of the comparison condition, In this case, The same, It is configured to input the qualified output 142 after the address replacement into the breakthrough count counter 135 of the one breakthrough address in each address group. In the indication, It is like changing the telescope previously described in Figure 7 to a parabolic telescope. And capture the burst output at the 1st breakthrough address. E.g, When 33 x 3 3 4 1000 pixels of the detection reference image A shown in Fig. 1 described so far is used as a reference for one coordinate of the image of the detection target image range, Just use the 1 coordinate as the center, And set the coordinate range of the XY axis 7000 pixels to the coordinate of the comparison range. Even in the case of being enlarged to twice the shape of the deformed image with rotation, It is also possible to detect an image by comparing the number of samples with the largest number of samples and the minimum number of samples. When the coordinates are changed or the portrait is reduced, etc. Since there will also be a lack of coordinates corresponding to the sample coordinates, therefore, It is only necessary to set the benchmark for the number of sample passes. This method, If it is exactly the same as the relative coordinates described in Figure 6, the method is compared with -42-201137651, Then the true sex system will be reduced, However, it is possible to perform an extremely high-speed image detection by 'recognizing the ability of the sample or setting the number appropriately. and then, If the data of the memory 132 is compared with the circuit 133, Judging whether or not the data is qualified or not, and comparing the size of the range having the brightness or the color, Not only a deformed portrait, Dependent on its definition method, It is also possible to detect similar portraits 〇 as above, In addition to the judgement of eligibility or not due to complete agreement, Also use size comparison, Range comparison, Memory bits are compared individually, Even able to perform 3 値 memory compared to Don’t Care, etc. And pass the pass or fail judgment, Can perform more effective information detection 〇 Usually, The size change of the image or the detection of the rotated image is applied. It requires a very large number of retrieval processing such as coordinate transformation. However, According to this way, Then only need to compare the number of samples, It is possible to detect a deformed image of the purpose (including an approximate image). In most cases, Almost all of them are as described above, as long as the deformed image or the center position of the similar image or the position of the center of gravity can be detected. but, Even when it is necessary to detect the enlargement or reduction of the image or the angle of rotation, As long as the data is added several times, it can be matched. In the case where there is a general need as described above, Once the image is stored, -43- 201137651 is detected in the range, After that, as long as it is as shown in Figure 10, For 2 4, 3, Where are the 4 diagonal coordinates of 5? As divided into 4 segments of detection range A, Split detection range B of 1 6 divisions for general division, To limit and detect the scope, In the case of 4 divisions, Just do it 16 times, In the case of 16 splits, Just do it 64 times, And a total of 80 times of data comparison, It is possible to grasp the aspect of the deformation of the portrait. usually, When such a deformed image in which the degree of deformation cannot be estimated is detected, It is necessary to speculate on the deformation of the image under consideration and to make a majority of the coordinate transformations. And pattern matching, Compared to the detection of this deformed image, The present invention enables unparalleled high speed pattern matching. If the segmentation range is subdivided, It is also possible to perform more accurate tests. The above is an example of this. but, By adding the minimum number of data comparisons in this way, It is also possible to detect complex images. In this case, A description will be given of an example in which a full-sample point 1 1 3 is set as a target in a wide range as a premise that the image has been expanded and reduced, and the possibility of rotation is assumed. but, It is also possible to individually specify a certain range for each sampling point 1 1 3 and perform a test. This method is used to supplement the uncertainty between the data and its address (location). When the same or similar data is continuously present, the system has important significance. If the definition of similar portraits is based on the location of the sample point 113 -44 - 201137651 or its data, It is possible to detect the same image widely, Approximate portrait, Deformed portrait, Similar to the image. The body of the invention is 121, In addition to the two detection methods described above for the coordinate shifting method (which makes the address coincide with the relative address to be compared) and the coordinate range mode (which allows the address to exist within the range of the coordinates to be compared), It is also possible to perform an address replacement application, for example, outside the range of coordinates to be compared, In any case, it can be realized only by the data setting of the address comparison data 127 of the address substitution circuit 134. Therefore, The system can be made into an integrated structure. By combining these image detection methods, It is able to perform more diverse image detection. In this case, In order to simplify the description, And for R, G, The method of combining the color data of B into the data of one address is explained. But it’s even if it’s R, G, The addresses of each of B are independent of each other and are compared. It can also be easily implemented. Memory 1 1 of the present invention Since it is a memory that can be constructed in parallel with the basic structure of the memory data, For example, in associative memory, etc. By means of the replacement of the address (address replacement), And a counter that counts the number of passes, And a very simple construction of a general priority encoder, therefore, The large-capacity system is also easy. Needless to say, Memory 1 21 of the present invention, Is a device, It will fundamentally solve the problem of the number of explorations (comparisons) based on the relationship between the data and the address of its data. As long as it is properly selected -45 - 201137651, the selected sample, It is possible to guarantee the minimum number of data comparisons with the number of samples as the upper limit. and, It can also be applied to the detection of portraits caused by the coordinates of similar features that have been clustered or to various other types of information detection. In the description so far, Although it is explained in the case of repeated filtering of information, but, Needless to say, It is also possible to set the filter to only one comparison, Single-shot information detection for 2 comparisons. As above, If the memory of the present invention having intelligent knowledge capable of self-information detection can be used, 1 2 1, In the case of information retrieval, Since the CPU or GPU only needs to assign input data and read the result, therefore, The system can greatly reduce the burden. also, Because it is capable of extremely high speed information detection, therefore, When the capacity of the memory is insufficient, Information can also be segmented and information detected. The data comparison processing time of each of the billions of the body 1 2 1 of the present invention, Since even if the processing time of the conservative estimation is assumed to be 1//second of each time, regardless of the size of the portrait, It is also possible to reliably detect the information of the destination in a period of several V seconds or even hundreds of microseconds. therefore, Even if it is detected by the information on the frame (frame) of the moving image, Or if the known information 1 〇 1 to be detected is a continuous large number of cases, It can also be widely used. of course, It is possible to perform a higher level of information detection by using a combination (usually) between the CPUs that access the cells 132 of the memory 121 of the present invention and perform successive processing. -46- 201137651 The above is a description of the present invention with the image information as the center. 'But the method of detecting information and the memory of the present invention 1 2 1 'for 1D information (sound, etc.) Or in the detection of information that is arranged as a multidimensional space, Also effective. [Embodiment 8] FIG. 1 2 (the third example of the address replacement means) is such that the information of the two-dimensional figure of FIG. 6 is expanded to X, Y, The case where the 3 axes of Z are arranged in the same arrangement in the 3D space or the similar arrangement is detected is possible. And of course, it can also be set to N-dimensional space. Specific examples of information detection, It will be described later. The Jiyue body 1 2 1 ' system with the information filtering detection function described so far can count the number of breakthroughs for all the information. And by the number of breakthroughs, the same information and similar information are continuously detected. But in the case where the number of memories is large, Or for the purpose of simplifying the circuit structure, The breakthrough counter 135 can also be OR gate 136, The number of circuits of the conditional gate 1 37 is reduced. [Embodiment 9] Fig. 1 3 (Reduction of information filtering detection circuit) memory 1 2 1 a, According to Figure 4, The basic concept of the memory 121 of the present invention shown in Fig. 5 is reduced for the number of circuits. The number of circuits after the break count counter 135 which is generally utilized as described so far, In the usual case, Is the number of -47-201137651 of the first breakthrough address that appears in one comparison (in the description so far, It is 2 million addresses and has 4096 resolutions. Average 48 8 addresses), Pay attention to this matter, And the number that will match this, For example, the number of circuits after the breakthrough counter 1 3 5, For example, one thousandth or two thousandth of the number of addresses of the memory 1 3 2 is reduced. In the picture, The system is reduced to the output of A~X. In this case, Just put the counter, It is assumed that the group break address of the address of each of the address groups illustrated in Fig. 7 can be memorized and the address is read out. Further, it is assumed that the break address can be read once at each of the counters 158, and when the configuration is such a configuration, The address replacement circuit 134 may be equipped with a simple address calculation processor or the like for address replacement. By using the arithmetic processor in such a way to increase the degree of freedom of address replacement (address replacement), The department can also expect information filtering with a more diverse approach. Suppose that when the number of breakthrough addresses is large and the overflow occurs, It is only necessary to set an alarm and change the sample for one comparison. also, As a method of reducing the number of circuits, It is also possible to use a circuit configuration in which double parallel logic calculation is switched to each memory of the memory and implemented. The ability to count all memory and count the number of passes, There is no limit to the number of occurrences of a breakthrough address. And ideally, but, Even so, the memory of the information filtering function is simplified. The memory 121a with information filtering detection function is simplified. It is also possible to reliably detect the purpose information by the method of information detection described so far. By setting the above general configuration, It is possible to increase the degree of freedom of the number of addresses of the memory of the memory 1 1 1 a of the present embodiment or the number of bits thereof (that is, the capacity of the memory). also, It can also be set to 3 値 memory as described above. Figure 1 4 (a commemorative example of the multiplexed data and the double parallel logic of the address), This is an example of multiplexing the double parallel logic calculations illustrated in Figure 4. As shown in Figure 14. The memory 1 2 1 b ' of this embodiment compares the comparison data used for comparison with the data of the memory with the data for comparison with the address, And a double-parallel pass or fail determination circuit, And a double parallel logic product calculus circuit, There are 2 groups each, And it becomes the composition of the logical product calculus and the logical OR (OR) calculation and output. 此种 This composition, The circuit of the Billion 1 1 1 shown in Fig. 5 can also be constructed for use. By setting this structure, It is possible to detect two patterns simultaneously. In this case, One of the examples of multiplexing, It can also be set as a combination of most of the two groups. The calculation can also be set to logical product (AND) or exclusive logic other than logical sum (OR) or any other logical calculus. -49- 201137651 This is done by multiplexing in response to the type or purpose of the information detected, And using a variety of logical calculus for double parallel logic calculations, It is able to conduct more high-level information detection. If it is the figure 4 described so far, Figure 14 and so on, And for the memory of the present invention 1 2 1 , 1 2 1 a, The composition of 1 2 1 b is summarized, Then it becomes: A memory with information filtering detection function, The memory, a memory that remembers information in each memory address and can read the information. Its characteristic is that The system has the following (1) to (4): (1) a first comparison data for juxtaposing data that is given from the outside and used for comparison with the data memorized in the memory, And the second comparison data for juxtaposing the addresses with the addresses of the memory objects, The comparison data of these are used as input means for input; And (2) by means of the first comparative data and by means of a material that is stored in the memory for juxtaposition and for determining whether to pass or not; And (3) by means of the second comparative data, and the addresses of the memory are compared in parallel with each other and the means for determining the eligibility; And (4) will be above (2), (3) The result of the judgment of the eligibility of both parties is a logical calculation method for the results of the logical calculation of the data and the conformity of the address at each address. and then, As an example of a circuit for filtering information of this memory into a simple circuit, It becomes a memory with information filtering detection function. among them, The aforementioned logical calculation of the memory with the information filtering detection function, Is a logical product (AND) calculus, and, The system has -50- 201137651 or less (1) to (3): (1) at the time of initial information detection, Judging whether the pass or fail is judged in parallel with the data of the memory by the first comparison data, And use at least one of the qualified memory addresses as a means of remembering the address; And (2) at the next information inspection, Judging whether the pass or fail is judged in parallel with the information of the company by the new first comparative data, And the qualified memory address, The address of the replacement is made by the address replacement means performed by the second comparative data; And (3) will break through the above (1), (2) The address of the logical product (AND) of the address is used as the means of output. and then, As shown in Figure 5, As shown in Figure 13 and the like, In order for this memory to continuously perform a double parallel logic calculation, It is a memory that performs logical product (A N D ) calculus with information filtering detection function. Among the above, the memory with the information filtering detection function, The system has the following (1) to (4): (1) a counter means for measuring the result of the comparison of the above-mentioned comparison data which is given a specific number of times for each memory address for each memory address; And (2) during the aforementioned initial information detection, The above counter of the qualified memory address is made a positive number of 1 (c〇unt up), And use this as a means of breaking through the address once mentioned above; And (3) the above-mentioned logical product (-51 - 201137651 AND ) calculation result in the above-mentioned 1st breakthrough address by the above-mentioned address replacement means when the above-mentioned next information detection is repeatedly given The cumulative counter is used as a means of breaking the address as N (more than 2 comparison times); And (4) the means of outputting the address of the N times of the above (3). and then, It is a memory with information filtering detection function. Wherein the above-mentioned means of replacing the address of the memory of the logical product (AND) calculation, The replacement of the address is performed in parallel by the entire range of addresses by the second comparison data. and then, It is a kind of information body with information filtering detection function. among them, The data of the second comparison data which is the memory of the logical product (AND) calculation described above and the address of the address of the symbolic body are juxtaposed with each other, Is the following (1), (2) One of the information detection uses When the aforementioned one breakthrough address is used as the reference address, And in the next time after the information detection, A memory address that is qualified and qualified in parallel with the data of the memory by the first comparison data, And the above reference address, The relative positions of the addresses of the two parties, And the comparative data used to determine whether (1) is consistent; (2) Whether there is comparative information in the scope. and then, This memory, It is possible to reduce the circuit used for filtering or to multi-parallel logic calculations. and, In the logic calculation, Not only a logical product or a logical sum, It is also possible to perform double parallel logic calculations of data and addresses for various logical calculations. the above, Ending the memory 1 2 1 of the present invention, 1 2 1 a, Description of 12 lb itself, the following, For one-dimensional, The multi-dimensional space information detection -52- 201137651 example is explained. [Embodiment 1 0] Fig. 1 5 (information detection example of address one-dimensional arrangement) ' For example, the horizontal axis corresponds to the address as the time axis, And the scene is pneumatic or stock price, Temperature and other data are displayed on the vertical axis. And it is information based on the known information given as a sample, Information Detector 'this kind of information detection' can be used to store information in a way that links memory addresses to timelines in a huge database of the past. , It is extremely simple to carry out. As another example of information detection using the time axis as an object, In the case of sound, The system can be based on the sampling time before compression and the sound data corresponding to the time or the sound data in each AAU (information decoding unit) of the compressed sound data. For information detection. As an example of the Kenji 1 1 1 of the present invention capable of comparing three pieces of data, If the band of the spectrum of human voices is classified and classified for each category, And set to 1 time, 1 bit of information, It is extremely simple to make a time series arrangement. If it is based on this, it is similar to the template source, etc. It is capable of extremely high speed sound recognition. And can be used in a variety of sound recognition fields. And the detection of such one-dimensional matching information, When DN A, which requires extensive information processing, is combined with the analysis of the combination of the four bases of the gene body at a high speed, It is extremely effective. -53- 201137651 Similarly, Analysis by the arrangement of the text columns, It is also the same. [Embodiment 1 1] Fig. 16 (Example of information detection of three-dimensional arrangement of addresses), It is an example of a case where the information listed in the three-dimensional space is detected. As shown in the figure, A schematic diagram of a specific pattern to be arranged in a three-dimensional space is detected by the same method as described so far. of course, Since the three-dimensional space system can represent the actual space that people exist, therefore, It is applicable to all three-dimensional information that can be used for the position of the three-dimensional space and the quantification of the data. It can also be expanded to include multidimensional information such as time and space information on the time axis. Information detection of such three-dimensional space, The system can be used in the analysis of the configuration relationship from the atomic or molecular level up to the universe. especially, Because it is capable of high-speed detection, therefore, It is also possible to target robots that require immediate processing. For example, a three-dimensional pattern recognition is generally performed by comparing a moving object or the like with a plurality of template images. Stereoscopic object recognition or object movement tracking, etc. Its use is unlimited. Whatever the case, Can be achieved by using the memory address with X, Y, The 3 axes of Z are added correspondingly to address and remember the data. And easy to implement, and, In addition to the same information, of course, It is also possible to perform similar information as described in the detection of an image or to detect similar information. -54- 201137651 The above description is based on a one-dimensional to multi-dimensional space. And for the morphological identification of the information that has been configured for the address or the information that can be assigned to the address, But as an example, The design of the address and the data is closely designed by classifying the address as a group of information groups that have been clustered, and so on. It can also be used for information detection using algorithms that have been performed so far. The above is a summary of the situation where one-dimensional to multi-dimensional information is detected. but, If it is for the memory 1 2 1 of the present invention, 1 2 1 a, The features of the role in 1 2 1 b are summarized, It is as follows. Using the memory 1 2 1 of the present invention, 1 2 1 a, 1 2 1 b The detection of information is based on the 1 site and its data as the unit of detection. And by setting the input method of the data, it becomes possible to approximate the information from the detection of the same information that is particularly strict and correct, Deformation information, a wide range of information, such as information, It is detected at high speed and surely. Due to the number of samples used for information detection or the selection method of sampling point 1 1 3, Both can be evaluated by statistical methods and can also be automated. therefore, Can prevent the waste of sampling numbers, And can make the detection time a reasonable time, also, The ability to set the sampling point 113 at each address (coordinate), Also one of the characteristics. Also get a mention, It is because when there is no information that becomes the object in the unknown information 102, The interruption of the test is also extremely fast. 〇 This test method, As long as the system meets certain conditions, There is no limit to the size (capacity) of each other's information. therefore, Can apply to all information from -55- 201137651, And can also cooperate with the memory 1 2 1 of the present invention, 121a, The capacity of 121b is calculated to divide the information. Even in the case of a combination of one-dimensional information and multi-dimensional information and other information, It is also possible to identify the pattern of the purpose. also, Another important feature of this approach, The system is: As long as the unknown information 102 is properly configured as an address and memorized in the memory of the present invention. 1 2 1 a, 1 2 1 b, Then except for the sampling point 1 1 3 It is not necessary to perform general information processing in other pattern recognition, Pre-processing of information such as feature extraction or categorization. and then, This way, At a point before the implementation of the information detection, such as the development of the algorithm, Also become an important feature, And, The detection method (setting of the comparison conditions) is also extremely simple. Therefore, The cut and try adjustments during system testing are also unnecessary. The information to be detected can be reliably detected by the set detection method. Therefore, Even if it is not an expert in information detection, It can also be widely used in various information detection applications by this method. In the description so far, It is based on the fact that the information is known as the first thing, and then the sample is taken to detect the information. And made a statement, but, According to the judgment or speculation of the person, setting the input data 125 and performing information analysis, It is also important. In this case, As a comparison of the initial input data 125, The system will have a significant impact on the results of the test, therefore, As long as -56- 201137651 is set as a comparison of plurals, Or, in the input data 125, it has a range to detect. And by means of the result of this comparison, the method of limiting the range is gradually gradually performed. Let's analyze it. This kind of analysis can be astronomical, meteorological, physical, Chemistry, The time for the analysis of information such as information forecasting in all areas of the economy and the labor involved have been greatly reduced. The memory of the invention with information filtering function 1 2 1 1 2 1 a, 1 2 1 b, It is to eliminate the processing of the memory of the fate of the von Neumann type computer. And let the memory itself have intelligent knowledge to conduct information detection. And the general knowledge of the memory so far has been greatly subverted. Therefore, It is not only able to use image recognition that has been studied and used from the past, Sound recognition, Ο C R text recognition, Full Text Search, Fingerprint authentication, Iridescent certification, The pattern recognition of robotic artificial intelligence, etc. can also carry out weather, Boom, share price, Molecular structure, DNA, Gene body, Analysis of texts, etc., and even the discovery of new information (information prediction), And can also make extensive use of social infrastructure, Industrial equipment, Industrial equipment, The home device has not been detected as an object of unknown information in the past. [Embodiment 12] The memory of the present invention having the information filtering function 1 2 1 1 2 1 a, The 1 2 1 b ' can also be used as an engine of artificial intelligence that requires detection of various kinds of information. -57- 201137651 For example, Can be utilized in extremely high level of knowledge processing, E.g, If the invention is based on the billion body 1 2 1 1 2 1 a, 1 2 1 b is regarded as the brain or cerebellum in the human brain, And further divided into the right brain or the left brain and the like, juxtaposed and hierarchically arranged in plural. And as a template to make the information of the object in advance, Information on character recognition, Information on text recognition, Information on sound recognition, Taste information, Various kinds of information such as tactile information, etc. And by comparing with the surrounding images or sounds given by the immediacy, and even the information of various sensors, The recognition ability of the person is generally recognized at the same time for various information. And from the information to choose the most appropriate action and implementation, Using the method, There are endless possibilities. Figure 17 (Example of highly knowledge processing using the memory of the present invention), In order to memorize the knowledge information described above in the memory 1 2 1 of the present invention, 1 2 1 a, An embodiment of knowledge processing in 1 2 1 b, and, For memory 121, based on the results of the most appropriate actions, 121a, 121b for update, Learning effects can also be easily achieved. [Industrial Applicability] The invention of the present invention 121, 121a, 121b, The system goes beyond the memory of the prior art that has been hitherto, Its use is extremely extensive. And to create a new trend of information processing. Memory 1 1 1 of the present invention, 1 2 1 a, 1 2 1 b, In addition to the associative memory, It can also be implemented as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array) as a general RAM or a ROM memory. -58- 201137651 and 'like the CPU of the unit base (ceU base), Directly into the CCD sensor, The memory i 2 1 of the present invention, 1 2 u, 121b is set to a memory structure of a unique structure. Or it is made up of a new form of semiconductor, Or give other functions, etc. Also for freedom, in the future, In addition to the above general memory caused by semiconductors, It is also possible to expect optical components or magnetic components that are being studied now, The memory 121 of the present invention caused by a new component such as Josephson 121a, 121b. [Simplified illustration of the drawing] [Fig. 1] Example of sampling points (Example 1). [Fig. 2] An information detecting embodiment (Example 2) caused by sampling points. [Fig. 3] An example of the evaluation method of the sampling point (Example 3). [Fig. 4] A commemorative example of the double parallel logic calculation of data and address. [Fig. 5] An example of a symbolic device having an information filtering detection function (Embodiment 4) « [Fig. 6] A first example (Example 5) of the address replacement means. FIG. 7 is a first schematic example of address replacement. FIG. 8 is a second schematic example of address replacement. [Fig. 9] A third exemplary example of address replacement. [Fig. 10] A concept of detection of an deformed image (Embodiment 6). [Fig. 1] A second example of the address replacement means (Embodiment 7). [Fig. 12] A third example of the address replacement means (Embodiment 8). -59 - 201137651 [Fig. 1 3] Example of reduction of information filtering detection circuit (Embodiment 9). [Fig. 14] A commemorative example of the multiplexed data and the double parallel logic of the address. [Fig. 15] An example of information detection in which the address is one-dimensionally arranged (Embodiment 10). [Fig. 16] An example of information detection for three-dimensional arrangement of addresses (Embodiment 1 1). [Fig. 17] An example of treatment using the high-knowledge knowledge of the genomic body of the present invention (Example 1 2). [Main component symbol description] 1 〇 1 : Known information 102 : Unknown information 1 0 3 : Address 1 1 1 : Pixel 1 1 3 : Sampling point 1 2 1. 1 2 1 a, 1 2 1 b : Memory with information filtering detection 122 : Address bus 123: Data bus 124: Output bus 125 : Input data 1 2 6 : Memory comparison data 1 2 7 : Address comparison data 128 : Reset signal 129 : Comparison counter 1 3 1 : Address Decoder -60- 201137651 1 3 2 : Memory 1 3 3 : Data comparison circuit 1 3 4 : Address replacement circuit 1 3 5 : Breakdown counter 136: Ο R gate 1 3 7 : Condition gate 1 3 8 : Breakthrough address output processing circuit 1 4 1 : Qualified output before address substitution 142 : Qualified output after address substitution 143 : Comparison number signal 144 : Cascade connection 1 5 8 : Group break counter -61 -