In this paper, we contribute two reconfigurable integer motion estimation (IME) architectures (namely RSADT and RPPSAD) based on adaptive algorithm. Firstly, based on the pixel difference analysis, the spatial redundancy is further exploited and three subsampling patterns are selected adaptively. Secondly, in order to keep full data reuse, we propose an architecture level data organization for RSADT architecture. For RPPSAD, we apply pixel classification and memory organization to keeps full data reuse. An interactive data loading scheme is proposed to reduce power dissipation. Experiments show that, with some extra hardware, our RSADT can averagely achieve 65.86% reduction in processing time; as for RPPSAD, it can save 25.4% to 39.8% power dissipation when processing typical HDTV720p sequences.