Data Compression for Video-Conferencing using Half tone and Wavelet Transform

Overhead of data transmission over internet is increasing exponentially every day. Optimization of natural bandwidth is the basic motive by compressing image data to the maximum extend. For the same objective, combination of lossy half tone and lossless Wavelet Transform techniques is proposed so as to obtain low-bit rate video data transmission. Decimal values of bitmapped image are to be converted into either 1 or 0 in half toning process that incur pictorial loss and gives 8:1 compression ratio (CR) irrespective of image. Wavelet Transform is applied on half tone image for higher compression for various levels. An experimental result shows the higher CR, minimum Mean Square Error (MSE). Ten sample images of different people captured by Nikon camera are used for experimentation. All images are bitmap (.BMP) 512 X 512 in size. The proposed technique can be used for video conferencing, storage of movies and CCTV footage etc.


INTRODUCTION
From last two decades Wavelet Transform has found enormous application in different areas like speech, computer graphics, signal, image processing and in medical field for DNA, ECG, protein, blood pressure, and heart rate analysis.Wavelet Transform overcomes the limitations of Fourier Transform as it cannot detect local properties as in [1].Hybrid Wavelet Transform using any two orthogonal transform can be used for higher image data compression with minimum loss using a set of complimentary wavelets, where comparison of DCT, DHT, DWT and Kekre transform is explained as in [2].The combination of Wavelet Transform with Modified-Run-Length-Coding (MRLC) along with new quantization technique is proposed for ECG data compression.This proposed method improves data compression by 13 % as in [3].
Generation of Wavelet Transform from any orthogonal transforms by contraction and translation infinite set of functions can be generated.Experimental results of original image with reconstructed image using orthogonal transforms Walsh and DCT with respect to their Wavelets are compared.Walsh Wavelet and DCT Wavelet results are better than Walsh and DCT as in [4].
Wavelet Transform for high resolution satellite imageries with lifting scheme is proposed that reduces computational time and resources with appreciable results as in [5].
Considering main three factors of high embedding capacity, imperceptibility and robustness effective stenography is explained with Walsh Wavelet and DCT Wavelet proven that are prone to filtering, noise, cropping and compression of an image as in [6].Spikes at different frequencies and amplitude using Wavelet Transform for Neural data compression from different channels are found to reconstruct unique signature and relate it some activities as in [7].Various orthogonal Wavelet transforms of Walsh, Cosine, Hartley, Kekre are used for image data compression and proved better results as compared to respective normal forms.70% to 90% is compressed by removing low energy coefficients in their respective Wavelet forms as in [8].
Section II explains about the half tone method and various half tone operators, Section III explains the Hybrid Algorithm of Half tone and Wavelet Transform, Section IV explains about experimental results and discussion.In section V paper conclusion and future scope is explained.

A. Neighbourhood Processing
Half toning is the process in which intensity and pattern of dot varies to simulate different shades.Half tone dots are produced by superimposing mask over the image.Half toning is the error diffusion process that results into noisy image.Half toning templates shown in fig.1a to fig.1d are used to convert continuous tone image into half tone image.These templates are rotated on continues tone image as neighborhood processing.
For the same objective of high image data compression and low-bit-rate data transmission to optimize bandwidth for video conferencing, other techniques are used as hybrid technique with half tone technique.Half toning is the lossy technique and gives 8:1 CR.Two-fold hybrid techniques are used for higher compression ratio with half tone.Half tone with Kekre's Fast Codebook Generation (KFCG) vector quantization technique is presented by Kekre et al as in [9].Lossy half tone with lossless Huffman coding technique is presented as in [10].Lossy half tone with lossless Run-Length-Encoding technique is presented as in [11].Importance of red plane from time complexity point of view is explained as in [12].For reconstruction of image from half tone image Inverse half toning algorithm as in [13] is described.Some other half toning operators are proposed with performance analysis as in [14].www.ijacsa.thesai.org

B. Quantization
As shown fig. 2 color image is split into three primary R-G-B planes and it posses gray levels from 0-255, representing each pixel by 8-bit.After half tone technique, quantization process is used to convert gray level into bi-level with loss as either 0 or 1 as in [13].

III. HYBRID HALF TONE WITH WAVELET ALGORITHM
As shown in fig.2, on each plane of half tone image Haar Wavelet transform is applied.Fig. 3 shows the working principle of Wavelet transform. Wavelet transform encoded data in its highest compressed form can be used for transmission on channel.
 At the receiving end inverse Wavelet transform is applied to decode image data so as to obtain half tone image.
 Inverse half toning algorithm is applied with concatenation of all the half tone planes to reconstruct of an image.

B. Compression Ratio (CR)
In first iteration of Wavelet transform image size of 512-by-512 in half tone form is converted into 256-by-256 as level-1 and referred as L1.At L1 Wavelet transform compresses data 50% to that of half tone image data as shown in fig.Fig. 13 shows the reconstructed images from different half tone operators and Wavelet Transform at levels from L1 to L4. Reconstructed image quality of Small operator is almost same to that of standard Floyd-Steinberg and Jarvis operators.As well as it gives same MSE and SSIM with reduced computational complexity [11].
Whereas South-East operator gives higher MSE and negligible poor in image quality as compared to standard operators.Fig. 13 shows the MSE increases and image quality decreases from L1 to L4.As shown in fig.3, in each level only Low-Low frequency component is used to take inverse Wavelet transform.Remaining Low-High, High-Low and High-High components are considered as matrix of zeros of the same size in the respective level.Image data can be compressed to the higher level of Wavelet Transform say 16X16 and 8X8, but the image reconstruction quality degrades a lot.Below this level reconstructed image quality degrades.Future scope to this paper is to convert Wavelet Transform domain image into desired number of non-overlapping blocks.Calculate energy of all the blocks and can eliminate the some lowest energy blocks based on threshold.Elimination of such non-overlapping blocks will increase the CR.In real-time processing, proposed algorithm takes more processing time as compared to the frame rate that required for smooth video-conferencing.As well as to develop an algorithm to preserve the features of Low-High, High-Low and High-High components of Wavelet transform domain image and can be added to Low-Low frequency component of Wavelet transform.

Figure 2 .
Figure 2. Block diagram of Half tone-Wavelet Transform Figure 3. Wavelet Transform Pyramid Fig. 5, 7, 9 and 11 shows the graphical representation of MSE between original images and inverse image for different half tone operators at L1, L2, L3 and L4.As well as fig.6, 8, 10 and 12 shows the SSIM between original image and inverse images for different half tone operators at L1, L2, L3 and L4.

Figure 6 .Figure 7 .Figure 8 .
Figure 6.SSIM between inverse and original images using different half tone operator and Wavelet Transorm at L1

Figure 9 .Figure 10 .
Figure 9. MSE between inverse and original images using different half tone operator and Wavelet Transorm at L3

Table - I
, III, V and Table-VII shows the MSE between original images and inverse image for different half tone operators at L1, L2, L3 and L4.As well as Table-II, IV, VI and Table-VIII shows the SSIM between original image and inverse images for different half tone operators at L1, L2, L3 and L4.

TABLE I .
MSE-BETWEEN INVERSE AND ORIGINAL IMAGE USING DIFFERENT HALF TONE OPERATORS WAVELET TRANSFORM AT-L1 S.N.

TABLE II .
SSIM-BETWEEN INVERSE AND ORIGINAL IMAGE USING DIFFERENT HALF TONE OPERATORS WAVELET TRANSFORM AT-L1

TABLE V .
MSE-BETWEEN INVERSE AND ORIGINAL IMAGE USING DIFFERENT HALF TONE OPERATORS WAVELET TRANSFORM AT-L3

TABLE VII .
MSE-BETWEEN INVERSE AND ORIGINAL IMAGE USING DIFFERENT HALF TONE OPERATORS WAVELET TRANSFORM AT-L4

TABLE VIII .
SSIM-BETWEEN INVERSE AND ORIGINAL IMAGE USING DIFFERENT HALF TONE OPERATORS WAVELET TRANSFORM AT-L4For low-bit rate video data transmission image data is compressed using combination of half tone and Wavelet Transform on ten different 512 by 512 bitmap images.Wavelet Transform is applied at different levels that converts image from 512 by 512 to 256 by 256 as level L1, 128 by 128 as L2, 64 by 64 as L3 and 32 by 32 as L4.
Figure 11.MSE between inverse and original images using different half tone operator and Wavelet Transorm at L4 Figure 12.SSIM between inverse and original images using different half tone operator and Wavelet Transorm at L4