CAJ | 학술논문

对残差信号用类高斯分布建模，通过分析回溯型自适应正交匹配追踪(backtracking-based adaptive orthogonal matching pursuit, BAOMP)算法的阈值选择方法与常规信号稀疏度方法的一致性和差异，提出一种改进的BAOMP算法。采用80-20准则判断信号的粗匹配状态，然后对后续匹配步骤引入可变步长阈值，实现选入原子集容量的精细调整，提高选入原子的正确匹配率，避免了残差信号的准周期性失配。实验结果表明，与BAOMP算法相比，在500次重复实验中，改进的BAOMP算法对高斯稀疏信号的精确重建概率提高17%～26%，对自然图像的精确重建概率提高70%以上。
대잔차신호용류고사분포건모，통과분석회소형자괄응정교필배추종(backtracking-based adaptive orthogonal matching pursuit, BAOMP)산법적역치선택방법여상규신호희소도방법적일치성화차이，제출일충개진적BAOMP산법。채용80-20준칙판단신호적조필배상태，연후대후속필배보취인입가변보장역치，실현선입원자집용량적정세조정，제고선입원자적정학필배솔，피면료잔차신호적준주기성실배。실험결과표명，여BAOMP산법상비，재500차중복실험중，개진적BAOMP산법대고사희소신호적정학중건개솔제고17%～26%，대자연도상적정학중건개솔제고70%이상。
This paper models residual signals with Gaussian-like distributions, based on which consistency between the Backtracking-based adaptive orthogonal matching pursuit (BAOMP) threshold and signal sparse-level is analyzed. An improved BAOMP (IBAOMP) method is thenproposed. Themethod estimates the preliminary matching state usingthe 80-20 rule, and introduces a threshold with variable step size to subtly adjust atom set to raise the correct rate of selected atoms and avoid quasi-periodic mismatches of residual signals. Simulation results of 500 tests show that the exact recovery probability of IBAOMP is 17%～26%higher than BAOMP for Gaussian sparse signals, and more than70%higher than BAOMP for natural images.