Chebiyyam Prabhakar

Underwater image denoising using adaptive wavelet subband thresholding

Recently, image denoising using the wavelet transform has been attracting much attention. Wavelet based approach provides a particularly useful method for image denoising when the preservation of image features in the scene is of importance. In this paper, we propose a novel denoising method for removing additive noise present in the underwater images. In addition to scattering and absorption effects, macroscopic floating particles producing images of the size of a pixel can be present as well due to sand raised by the motion of a diver, or small plankton particles. These particles are part of the scene, but cause generally unwanted signal. We see them as an additive noise. The problems it causes in feature extraction. In the proposed denoising method, first we use homomorphic filtering for correcting non uniform illumination, then we apply anisotropic filtering for smoothing. After smoothing, we apply adaptive wavelet subband thresholding with Modified BayesShrink function. We compared and evaluated the proposed denoising method based on the Peak Signal to Noise Ratio (PSNR). The experimental result shows that the proposed method yields superior result for underwater noisy images compared to other denoising techniques.

LBP-SURF descriptor with color invariant and texture based features for underwater images

In this paper, we introduce LBP-SURF, a local image descriptor for underwater environment, which is very efficient to extract color invariant and texture based features of underwater images. The current state-of-the-art feature descriptors viz. SIFT, SURF, DAISY, GLOH and variants are well known techniques for detecting and describing features of objects captured in the out-of-water environment. These standard descriptors have been proven to be the most robust to geometric variations. Nearly, all these geometrical invariant approaches avoid dealing with color images due to the color constancy problem. In underwater images, variation in color is very high compared to variations in geometrical properties due to propagation properties of light. The literature survey reveals that, the texture parameters that remain constant for the scene patch for the whole underwater image sequence. This motivated us to consider texture and color invariant features of underwater images, instead of using the gray-based geometrical invariant features. We normalize the color image using comprehensive color image normalization method to render the color values changed by the various radiometric factors of underwater environment. Our method uses Speeded Up Robust Features (SURF) to detect interest points from the normalized image. The texture features are extracted, and description is stored using Center-Symmetric Local Binary Patterns (CS-LBP) descriptor. The combination of SURF and CS-LBP, called LBP-SURF is evaluated extensively to verify its effectiveness with datasets acquired in underwater environment.

Color Invariant Feature Detection and Matching in Underwater Stereo Images

In this paper, we present an approach to find correspondences in underwater stereo images based on detection and matching of feature points, which are invariant to photometric variations. In underwater environment, the problem of finding correspondences in stereo images is specific step in order to estimate the motion of an underwater vehicle. The current state-of-the-art feature detectors have been proven to be the most robust to geometric variations and avoid dealing with color images due to color constancy problem. The propagation property of light in the underwater causes variations in color information between two underwater images taken under same imaging conditions. To render the color values changed by the various radiometric factors of underwater environment, we use comprehensive color image normalization method to normalize the color image. Our technique uses SIFT to detect interest points from the normalized image. In order to establish correspondences between images, we use window-based correlation measure instead of feature-based correlation techniques. The underwater images are low contrast in nature and lack of image features cause the feature-based techniques matching procedure to fail. Our approach is evaluated extensively to verify its effectiveness with data sets acquired in underwater environment. A new approach based on color invariant feature detection and window-based correlation matching significantly improves the matching reliability.

PARALLAX EFFECT FREE MOSAICING OF UNDERWATER VIDEO SEQUENCE BASED ON TEXTURE FEATURES

In this paper, we present feature-based technique for construction of mosaic image from underwater video sequence, which suffers from parallax distortion due to propagation properties of light in the underwater environment. The most of the available mosaic tools and underwater image mosaicing techniques yields final result with some artifacts such as blurring, ghosting and seam due to presence of parallax in the input images. The removal of parallax from input images may not reduce its effects instead it must be corrected in successive steps of mosaicing. Thus, our approach minimizes the parallax effects by adopting an efficient local alignment technique after global registration. We extract texture features using Centre Symmetric Local Binary Pattern (CS-LBP) descriptor in order to find feature correspondences, which are used further for estimation of homography through RANSAC. In order to increase the accuracy of global registration, we perform preprocessing such as colour alignment between two selected frames based on colour distribution adjustment. Because of existence of 100% overlap in consecutive frames of underwater video, we select frames with minimum overlap based on mutual offset in order to reduce the computation cost during mosaicing. Our approach minimizes the parallax effects considerably in final mosaic constructed using our own underwater video sequences.

Extraction of Texture Based Features of Underwater Images Using RLBP Descriptor

In this paper, we present an approach for extraction of texture features of underwater images using Robust Local Binary Pattern (RLBP) descriptor. The literature survey reveals that the texture parameters that remain constant for the scene patch for the whole underwater image sequence. Therefore, we proposed technique to extract the texture features and these features can be used for object recognition and tracking. The underwater images suffer from image blurring and low contrast and performance of feature extractors is very less if we employ directly. Thus, we propose a novel image enhancement technique which is combination of different individual filters such as homomorphic filtering, curvelet denoising and LBP based Diffusion. We employ DoG based feature detector, for each detected interest point, the texture description is extracted using RLBP feature descriptor. The proposed feature extraction technique is compared and evaluated extensively with well known feature extractors using datasets acquired in underwater environment.

Quality assessment of stereo images using reduced reference based on saliency region

In this paper, we present a novel technique to asses the quality of the stereo images in its reduced reference based on saliency region. One of the important factors that influence the accuracy of text information extraction algorithms is quality of the acquired images. Therefore, we developed technique to measure quality of stereo images which assists in selection of high quality ones in order to increase the performance of text information extraction algorithm. The planar surfaces become the salient regions in Text Information Extraction (TIE), because text information is always contained in planar surfaces as evidenced in day to day encounters. The proposed technique involves the extraction of planar surfaces from both left and right frames based on estimation of spatial gradient in disparity space, and later on performs Reduced Reference (RR) quality assessment of frames using the extracted saliency region. The various experiments are conducted in order to evaluate the performance of proposed method using stereo video sequences acquired using digital video cameras and mobile phone devices.

Localization of Curved Scene Text in Video based on Fuzzy Curve Tracing of Extracted Planar Surface

In this paper, we present a novel technique to localize curved multi-script text contained in natural scene video based on Fuzzy Curve Tracing (FCT) of extracted planar surface. In order to read and interpret easily, text information is usually written on planar surfaces, for instance billboards, walls of buildings, road-signs and banners. This motivated us to detect planar surfaces by fitting a planar model, that is constructed using Random Sample Consensus (RANSAC). It is assumed that the detected planar surface contains text and is segmental from background using Graph Cuts through Markov Random Field (MRF) labeling of pixels belongs to planar surface. Within the extracted planar surface, the curved text is detected using fuzzy curve tracing, which traces and generates curve path of the text by establishing spatial relations among the cluster centers identified through fuzzy c-means clustering of character regions. Finally, curved text is localized by identifying character regions wherever generated curve path pass through it. The experimental results are evaluated for text localization using recall, precision and f-measure. Based on these metrics result, its incontestible that the projected technique outperforms the popular existing methods.

Stereo Correspondence for Underwater Video Sequence Using Graph Cuts

In this paper, we introduce stereo correspondence method for underwater video sequence using Graph Cuts. The propagation property of light in the underwater causes variations in color information between two underwater video frames taken under same imaging conditions. To render the color values changed by the propagation property of light in the underwater environment, we use Markov Random Fields - Belief Propagation (MRF-BP) based approach for color correction. The conventional window-based correlation methods are often employed to estimate the disparity between the image pair, but these techniques are sensitive to illuminative variations, leads to fattening effect at the object boundaries and relatively lower performance in the featureless regions. Therefore, we employ energy minimization method such as Graph Cuts for the pair of color corrected underwater video frames to estimate disparity map. We compared and evaluated our approach qualitatively with well known window-based stereo correspondence techniques for the captured underwater video test frames. The experimental result reveals that our approach yields a visually suitable dense disparity map for the captured underwater video test frames compared to a window-based stereo correspondence techniques.