【摘要】根據(jù)對(duì)向傳播網(wǎng)絡(luò)適于模式分類(lèi)的特性，提出了基于對(duì)向傳播網(wǎng)絡(luò)的人臉識(shí)別方法。同時(shí)，為了克服對(duì)向傳播網(wǎng)絡(luò)在訓(xùn)練過(guò)程中的不穩(wěn)定性，改進(jìn)了對(duì)向傳播網(wǎng)絡(luò)的學(xué)習(xí)算法，并且以國(guó)際通用的ORL人臉庫(kù)為對(duì)象進(jìn)行實(shí)驗(yàn)研究。結(jié)果表明，選擇合適的參數(shù)，就可以達(dá)到理想的識(shí)別效果。證明了基于對(duì)向傳播網(wǎng)絡(luò)的人臉識(shí)別方法是有效可行的。
關(guān) 鍵 詞 神經(jīng)網(wǎng)絡(luò); 對(duì)向傳播網(wǎng)絡(luò); 人臉識(shí)別; 模式分類(lèi)

It is well known that, counter propagation network is fitful for pattern classification. According to such attribute, we present a face recognition algorithm based on counter propagation network. Meanwhile, we modify the learning method of counter propagation network in case of its instability during the training phase, and do some experiments with ORL face library that is international universal. Experiments show that we can achieve sound result by choosing parameters appropriately. So, the recognition algorithm based on counter propagation network is effective and practicable.