Reservoir discontinuity is a practical representation of reservoir heterogeneity, which leads to the nonuniform flow of hydrocarbons during production and increases the difficulties of producing the remaining oil in clastic reservoirs. For offshore oilfields, reservoir discontinuity analysis will have to rely on seismic data due to their sparse well distribution. However, traditional discontinuity detection methods are restricted to large-scale discontinuities (such as faults). It is difficult to detect small-scale discontinuities in seismic data such as interfaces of sands. To interpret small-scale discontinuities below the seismic resolution, we have adopted a direction-adaptive mathematical morphology gradient algorithm that can detect the boundaries of reservoir architectures in different directions on horizon-based seismic attributes. We elaborated a hierarchy of fluvial reservoir architectures and classified compound sandstones as different architectural elements according to the deposition period. Forward models were designed to analyze the seismic responses of the architectures. The results demonstrate that amplitude-related seismic attributes could be sensitive to different reservoir architectures below the seismic resolution. The amplitude-related seismic attributes can be extracted through a time window guided by the horizons. The horizon-based attributes can be treated as digital images, which may contain the information of compound sandstones with hidden discontinuity boundaries. Then, our algorithm is applied to detect discontinuity boundaries based on this extracted attribute. The application to field data in the Bohai Bay demonstrates that our algorithm can detect the discontinuity boundaries inside the compound sandstone, which is less than the seismic resolution. This method could enhance our capability to visualize and understand the complexity of reservoir heterogeneity.