关键词: 尺度不变特征转换（SIFT）, 高斯滤波, Top-hat变换, 高斯差分, 描述子

Abstract:

In order to solve the problem of poor performance of feature point extraction algorithm in uneven illumination, an improved SIFT（Scale Invariant Feature Transform） algorithm is proposed to suppress the influence of uneven illumination. Firstly, in the scale-space construction, the input image is filtered by Gaussian high-pass filtering in frequency domain to filter the light components, and combined with Top-hat transform, Gaussian filter parameter selection is weakened, then a pyramid of difference of Gaussian is constructed by Gaussian convolution which based on light filtering and parameter weakening. Finally, SIFT algorithm is fused to generate GT-SIFT descriptor with good illumination invariance for feature point extraction and matching. The experimental results show that compared with the traditional algorithms, the improved algorithm under the condition of uneven illumination has better robustness, image feature point extraction and matching effect is better.

Key words: Scale Invariant Feature Transform（SIFT）, Gaussian filter, Top-hat transform, difference of Gaussian, descriptor