Video streaming applications are increasingly popular but suffer from quality variations in best effort transmission networks. The recent DASH standard allows to dynamically adapt the transmitted rate to the network conditions under the control of the receiving client to achieve the best user experience. Naturally, the final video quality is also critically determined by the compression efficiency of the adopted coding solution. Video coding standards such as H.264/AVC and HEVC are based on a complexity allocation paradigm where complex encoders serve simpler decoders. However, some application environments can afford more complex and thus more intelligent decoders to boost the final quality. In this context, this paper proposes a decoder quality boosting algorithm where any standard compliant bitstreams are decoded with better quality in the lower rate/quality periods that succeed to higher rate/quality periods due to network bandwidth variations. The quality boosting results from a machine learning based solution, which is able to reuse already decoded frames from the previous higher quality segment to improve the lower quality segment. Thus, by increasing the decoder complexity it is possible to increase the quality of experience in DASH based streaming frameworks. Experimental results using the HEVC video coding standard show that significant quality gains may be obtained, notably achieving a graceful quality degradation in the rate dropping transitions.