This paper presents a non-iterative deep learning approach to compressive sensing (CS) image reconstruction using a convolutional autoencoder and a residual learning network. An efficient measurement design is proposed in order to enable training of the compressive sensing models on normalized and mean-centred measurements, along with a practical network initialization method based on principal component analysis (PCA). Finally, perceptual residual learning is proposed in order to obtain semantically informative image reconstructions along with high pixel-wise reconstruction accuracy at low measurement rates.
Pub. online:1 Jan 2010Type:Research ArticleOpen Access
Volume 21, Issue 3 (2010), pp. 409–424
The paper addresses the over-saturated protein spot detection and extraction problem in two-dimensional electrophoresis gel images. The effective technique for detection and reconstruction of over-saturated protein spots is proposed. The paper presents: an algorithm of the median filter mask adaptation for initial filtering of gel image; the models of over-saturation used for gel image analysis; several models of protein spots used for reconstruction; technique of the automatic over-saturated protein spot search and reconstruction. Experimental investigation confirms that proposed search technique lets to find up to 96% of over-saturated protein spots. Moreover the proposed flexible protein spot shape models for reconstruction are faster and more accurate in comparison to the flexible diffusion model.